1 //===- BugReporter.cpp - Generate PathDiagnostics for bugs ----------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This file defines BugReporter, a utility class for generating 10 // PathDiagnostics. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h" 15 #include "clang/AST/Decl.h" 16 #include "clang/AST/DeclBase.h" 17 #include "clang/AST/DeclObjC.h" 18 #include "clang/AST/Expr.h" 19 #include "clang/AST/ExprCXX.h" 20 #include "clang/AST/ParentMap.h" 21 #include "clang/AST/Stmt.h" 22 #include "clang/AST/StmtCXX.h" 23 #include "clang/AST/StmtObjC.h" 24 #include "clang/Analysis/AnalysisDeclContext.h" 25 #include "clang/Analysis/CFG.h" 26 #include "clang/Analysis/CFGStmtMap.h" 27 #include "clang/Analysis/PathDiagnostic.h" 28 #include "clang/Analysis/ProgramPoint.h" 29 #include "clang/Basic/LLVM.h" 30 #include "clang/Basic/SourceLocation.h" 31 #include "clang/Basic/SourceManager.h" 32 #include "clang/StaticAnalyzer/Core/AnalyzerOptions.h" 33 #include "clang/StaticAnalyzer/Core/BugReporter/BugReporterVisitors.h" 34 #include "clang/StaticAnalyzer/Core/BugReporter/BugType.h" 35 #include "clang/StaticAnalyzer/Core/Checker.h" 36 #include "clang/StaticAnalyzer/Core/CheckerManager.h" 37 #include "clang/StaticAnalyzer/Core/CheckerRegistryData.h" 38 #include "clang/StaticAnalyzer/Core/PathSensitive/ExplodedGraph.h" 39 #include "clang/StaticAnalyzer/Core/PathSensitive/ExprEngine.h" 40 #include "clang/StaticAnalyzer/Core/PathSensitive/MemRegion.h" 41 #include "clang/StaticAnalyzer/Core/PathSensitive/ProgramState.h" 42 #include "clang/StaticAnalyzer/Core/PathSensitive/SMTConv.h" 43 #include "clang/StaticAnalyzer/Core/PathSensitive/SVals.h" 44 #include "clang/StaticAnalyzer/Core/PathSensitive/SymbolManager.h" 45 #include "llvm/ADT/ArrayRef.h" 46 #include "llvm/ADT/DenseMap.h" 47 #include "llvm/ADT/DenseSet.h" 48 #include "llvm/ADT/FoldingSet.h" 49 #include "llvm/ADT/STLExtras.h" 50 #include "llvm/ADT/SmallPtrSet.h" 51 #include "llvm/ADT/SmallString.h" 52 #include "llvm/ADT/SmallVector.h" 53 #include "llvm/ADT/Statistic.h" 54 #include "llvm/ADT/StringExtras.h" 55 #include "llvm/ADT/StringRef.h" 56 #include "llvm/ADT/iterator_range.h" 57 #include "llvm/Support/Casting.h" 58 #include "llvm/Support/Compiler.h" 59 #include "llvm/Support/ErrorHandling.h" 60 #include "llvm/Support/MemoryBuffer.h" 61 #include "llvm/Support/raw_ostream.h" 62 #include <algorithm> 63 #include <cassert> 64 #include <cstddef> 65 #include <iterator> 66 #include <memory> 67 #include <optional> 68 #include <queue> 69 #include <string> 70 #include <tuple> 71 #include <utility> 72 #include <vector> 73 74 using namespace clang; 75 using namespace ento; 76 using namespace llvm; 77 78 #define DEBUG_TYPE "BugReporter" 79 80 STATISTIC(MaxBugClassSize, 81 "The maximum number of bug reports in the same equivalence class"); 82 STATISTIC(MaxValidBugClassSize, 83 "The maximum number of bug reports in the same equivalence class " 84 "where at least one report is valid (not suppressed)"); 85 86 BugReporterVisitor::~BugReporterVisitor() = default; 87 88 void BugReporterContext::anchor() {} 89 90 //===----------------------------------------------------------------------===// 91 // PathDiagnosticBuilder and its associated routines and helper objects. 92 //===----------------------------------------------------------------------===// 93 94 namespace { 95 96 /// A (CallPiece, node assiciated with its CallEnter) pair. 97 using CallWithEntry = 98 std::pair<PathDiagnosticCallPiece *, const ExplodedNode *>; 99 using CallWithEntryStack = SmallVector<CallWithEntry, 6>; 100 101 /// Map from each node to the diagnostic pieces visitors emit for them. 102 using VisitorsDiagnosticsTy = 103 llvm::DenseMap<const ExplodedNode *, std::vector<PathDiagnosticPieceRef>>; 104 105 /// A map from PathDiagnosticPiece to the LocationContext of the inlined 106 /// function call it represents. 107 using LocationContextMap = 108 llvm::DenseMap<const PathPieces *, const LocationContext *>; 109 110 /// A helper class that contains everything needed to construct a 111 /// PathDiagnostic object. It does no much more then providing convenient 112 /// getters and some well placed asserts for extra security. 113 class PathDiagnosticConstruct { 114 /// The consumer we're constructing the bug report for. 115 const PathDiagnosticConsumer *Consumer; 116 /// Our current position in the bug path, which is owned by 117 /// PathDiagnosticBuilder. 118 const ExplodedNode *CurrentNode; 119 /// A mapping from parts of the bug path (for example, a function call, which 120 /// would span backwards from a CallExit to a CallEnter with the nodes in 121 /// between them) with the location contexts it is associated with. 122 LocationContextMap LCM; 123 const SourceManager &SM; 124 125 public: 126 /// We keep stack of calls to functions as we're ascending the bug path. 127 /// TODO: PathDiagnostic has a stack doing the same thing, shouldn't we use 128 /// that instead? 129 CallWithEntryStack CallStack; 130 /// The bug report we're constructing. For ease of use, this field is kept 131 /// public, though some "shortcut" getters are provided for commonly used 132 /// methods of PathDiagnostic. 133 std::unique_ptr<PathDiagnostic> PD; 134 135 public: 136 PathDiagnosticConstruct(const PathDiagnosticConsumer *PDC, 137 const ExplodedNode *ErrorNode, 138 const PathSensitiveBugReport *R); 139 140 /// \returns the location context associated with the current position in the 141 /// bug path. 142 const LocationContext *getCurrLocationContext() const { 143 assert(CurrentNode && "Already reached the root!"); 144 return CurrentNode->getLocationContext(); 145 } 146 147 /// Same as getCurrLocationContext (they should always return the same 148 /// location context), but works after reaching the root of the bug path as 149 /// well. 150 const LocationContext *getLocationContextForActivePath() const { 151 return LCM.find(&PD->getActivePath())->getSecond(); 152 } 153 154 const ExplodedNode *getCurrentNode() const { return CurrentNode; } 155 156 /// Steps the current node to its predecessor. 157 /// \returns whether we reached the root of the bug path. 158 bool ascendToPrevNode() { 159 CurrentNode = CurrentNode->getFirstPred(); 160 return static_cast<bool>(CurrentNode); 161 } 162 163 const ParentMap &getParentMap() const { 164 return getCurrLocationContext()->getParentMap(); 165 } 166 167 const SourceManager &getSourceManager() const { return SM; } 168 169 const Stmt *getParent(const Stmt *S) const { 170 return getParentMap().getParent(S); 171 } 172 173 void updateLocCtxMap(const PathPieces *Path, const LocationContext *LC) { 174 assert(Path && LC); 175 LCM[Path] = LC; 176 } 177 178 const LocationContext *getLocationContextFor(const PathPieces *Path) const { 179 assert(LCM.count(Path) && 180 "Failed to find the context associated with these pieces!"); 181 return LCM.find(Path)->getSecond(); 182 } 183 184 bool isInLocCtxMap(const PathPieces *Path) const { return LCM.count(Path); } 185 186 PathPieces &getActivePath() { return PD->getActivePath(); } 187 PathPieces &getMutablePieces() { return PD->getMutablePieces(); } 188 189 bool shouldAddPathEdges() const { return Consumer->shouldAddPathEdges(); } 190 bool shouldAddControlNotes() const { 191 return Consumer->shouldAddControlNotes(); 192 } 193 bool shouldGenerateDiagnostics() const { 194 return Consumer->shouldGenerateDiagnostics(); 195 } 196 bool supportsLogicalOpControlFlow() const { 197 return Consumer->supportsLogicalOpControlFlow(); 198 } 199 }; 200 201 /// Contains every contextual information needed for constructing a 202 /// PathDiagnostic object for a given bug report. This class and its fields are 203 /// immutable, and passes a BugReportConstruct object around during the 204 /// construction. 205 class PathDiagnosticBuilder : public BugReporterContext { 206 /// A linear path from the error node to the root. 207 std::unique_ptr<const ExplodedGraph> BugPath; 208 /// The bug report we're describing. Visitors create their diagnostics with 209 /// them being the last entities being able to modify it (for example, 210 /// changing interestingness here would cause inconsistencies as to how this 211 /// file and visitors construct diagnostics), hence its const. 212 const PathSensitiveBugReport *R; 213 /// The leaf of the bug path. This isn't the same as the bug reports error 214 /// node, which refers to the *original* graph, not the bug path. 215 const ExplodedNode *const ErrorNode; 216 /// The diagnostic pieces visitors emitted, which is expected to be collected 217 /// by the time this builder is constructed. 218 std::unique_ptr<const VisitorsDiagnosticsTy> VisitorsDiagnostics; 219 220 public: 221 /// Find a non-invalidated report for a given equivalence class, and returns 222 /// a PathDiagnosticBuilder able to construct bug reports for different 223 /// consumers. Returns std::nullopt if no valid report is found. 224 static std::optional<PathDiagnosticBuilder> 225 findValidReport(ArrayRef<PathSensitiveBugReport *> &bugReports, 226 PathSensitiveBugReporter &Reporter); 227 228 PathDiagnosticBuilder( 229 BugReporterContext BRC, std::unique_ptr<ExplodedGraph> BugPath, 230 PathSensitiveBugReport *r, const ExplodedNode *ErrorNode, 231 std::unique_ptr<VisitorsDiagnosticsTy> VisitorsDiagnostics); 232 233 /// This function is responsible for generating diagnostic pieces that are 234 /// *not* provided by bug report visitors. 235 /// These diagnostics may differ depending on the consumer's settings, 236 /// and are therefore constructed separately for each consumer. 237 /// 238 /// There are two path diagnostics generation modes: with adding edges (used 239 /// for plists) and without (used for HTML and text). When edges are added, 240 /// the path is modified to insert artificially generated edges. 241 /// Otherwise, more detailed diagnostics is emitted for block edges, 242 /// explaining the transitions in words. 243 std::unique_ptr<PathDiagnostic> 244 generate(const PathDiagnosticConsumer *PDC) const; 245 246 private: 247 void updateStackPiecesWithMessage(PathDiagnosticPieceRef P, 248 const CallWithEntryStack &CallStack) const; 249 void generatePathDiagnosticsForNode(PathDiagnosticConstruct &C, 250 PathDiagnosticLocation &PrevLoc) const; 251 252 void generateMinimalDiagForBlockEdge(PathDiagnosticConstruct &C, 253 BlockEdge BE) const; 254 255 PathDiagnosticPieceRef 256 generateDiagForGotoOP(const PathDiagnosticConstruct &C, const Stmt *S, 257 PathDiagnosticLocation &Start) const; 258 259 PathDiagnosticPieceRef 260 generateDiagForSwitchOP(const PathDiagnosticConstruct &C, const CFGBlock *Dst, 261 PathDiagnosticLocation &Start) const; 262 263 PathDiagnosticPieceRef 264 generateDiagForBinaryOP(const PathDiagnosticConstruct &C, const Stmt *T, 265 const CFGBlock *Src, const CFGBlock *DstC) const; 266 267 PathDiagnosticLocation 268 ExecutionContinues(const PathDiagnosticConstruct &C) const; 269 270 PathDiagnosticLocation 271 ExecutionContinues(llvm::raw_string_ostream &os, 272 const PathDiagnosticConstruct &C) const; 273 274 const PathSensitiveBugReport *getBugReport() const { return R; } 275 }; 276 277 } // namespace 278 279 //===----------------------------------------------------------------------===// 280 // Base implementation of stack hint generators. 281 //===----------------------------------------------------------------------===// 282 283 StackHintGenerator::~StackHintGenerator() = default; 284 285 std::string StackHintGeneratorForSymbol::getMessage(const ExplodedNode *N){ 286 if (!N) 287 return getMessageForSymbolNotFound(); 288 289 ProgramPoint P = N->getLocation(); 290 CallExitEnd CExit = P.castAs<CallExitEnd>(); 291 292 // FIXME: Use CallEvent to abstract this over all calls. 293 const Stmt *CallSite = CExit.getCalleeContext()->getCallSite(); 294 const auto *CE = dyn_cast_or_null<CallExpr>(CallSite); 295 if (!CE) 296 return {}; 297 298 // Check if one of the parameters are set to the interesting symbol. 299 unsigned ArgIndex = 0; 300 for (CallExpr::const_arg_iterator I = CE->arg_begin(), 301 E = CE->arg_end(); I != E; ++I, ++ArgIndex){ 302 SVal SV = N->getSVal(*I); 303 304 // Check if the variable corresponding to the symbol is passed by value. 305 SymbolRef AS = SV.getAsLocSymbol(); 306 if (AS == Sym) { 307 return getMessageForArg(*I, ArgIndex); 308 } 309 310 // Check if the parameter is a pointer to the symbol. 311 if (std::optional<loc::MemRegionVal> Reg = SV.getAs<loc::MemRegionVal>()) { 312 // Do not attempt to dereference void*. 313 if ((*I)->getType()->isVoidPointerType()) 314 continue; 315 SVal PSV = N->getState()->getSVal(Reg->getRegion()); 316 SymbolRef AS = PSV.getAsLocSymbol(); 317 if (AS == Sym) { 318 return getMessageForArg(*I, ArgIndex); 319 } 320 } 321 } 322 323 // Check if we are returning the interesting symbol. 324 SVal SV = N->getSVal(CE); 325 SymbolRef RetSym = SV.getAsLocSymbol(); 326 if (RetSym == Sym) { 327 return getMessageForReturn(CE); 328 } 329 330 return getMessageForSymbolNotFound(); 331 } 332 333 std::string StackHintGeneratorForSymbol::getMessageForArg(const Expr *ArgE, 334 unsigned ArgIndex) { 335 // Printed parameters start at 1, not 0. 336 ++ArgIndex; 337 338 return (llvm::Twine(Msg) + " via " + std::to_string(ArgIndex) + 339 llvm::getOrdinalSuffix(ArgIndex) + " parameter").str(); 340 } 341 342 //===----------------------------------------------------------------------===// 343 // Diagnostic cleanup. 344 //===----------------------------------------------------------------------===// 345 346 static PathDiagnosticEventPiece * 347 eventsDescribeSameCondition(PathDiagnosticEventPiece *X, 348 PathDiagnosticEventPiece *Y) { 349 // Prefer diagnostics that come from ConditionBRVisitor over 350 // those that came from TrackConstraintBRVisitor, 351 // unless the one from ConditionBRVisitor is 352 // its generic fallback diagnostic. 353 const void *tagPreferred = ConditionBRVisitor::getTag(); 354 const void *tagLesser = TrackConstraintBRVisitor::getTag(); 355 356 if (X->getLocation() != Y->getLocation()) 357 return nullptr; 358 359 if (X->getTag() == tagPreferred && Y->getTag() == tagLesser) 360 return ConditionBRVisitor::isPieceMessageGeneric(X) ? Y : X; 361 362 if (Y->getTag() == tagPreferred && X->getTag() == tagLesser) 363 return ConditionBRVisitor::isPieceMessageGeneric(Y) ? X : Y; 364 365 return nullptr; 366 } 367 368 /// An optimization pass over PathPieces that removes redundant diagnostics 369 /// generated by both ConditionBRVisitor and TrackConstraintBRVisitor. Both 370 /// BugReporterVisitors use different methods to generate diagnostics, with 371 /// one capable of emitting diagnostics in some cases but not in others. This 372 /// can lead to redundant diagnostic pieces at the same point in a path. 373 static void removeRedundantMsgs(PathPieces &path) { 374 unsigned N = path.size(); 375 if (N < 2) 376 return; 377 // NOTE: this loop intentionally is not using an iterator. Instead, we 378 // are streaming the path and modifying it in place. This is done by 379 // grabbing the front, processing it, and if we decide to keep it append 380 // it to the end of the path. The entire path is processed in this way. 381 for (unsigned i = 0; i < N; ++i) { 382 auto piece = std::move(path.front()); 383 path.pop_front(); 384 385 switch (piece->getKind()) { 386 case PathDiagnosticPiece::Call: 387 removeRedundantMsgs(cast<PathDiagnosticCallPiece>(*piece).path); 388 break; 389 case PathDiagnosticPiece::Macro: 390 removeRedundantMsgs(cast<PathDiagnosticMacroPiece>(*piece).subPieces); 391 break; 392 case PathDiagnosticPiece::Event: { 393 if (i == N-1) 394 break; 395 396 if (auto *nextEvent = 397 dyn_cast<PathDiagnosticEventPiece>(path.front().get())) { 398 auto *event = cast<PathDiagnosticEventPiece>(piece.get()); 399 // Check to see if we should keep one of the two pieces. If we 400 // come up with a preference, record which piece to keep, and consume 401 // another piece from the path. 402 if (auto *pieceToKeep = 403 eventsDescribeSameCondition(event, nextEvent)) { 404 piece = std::move(pieceToKeep == event ? piece : path.front()); 405 path.pop_front(); 406 ++i; 407 } 408 } 409 break; 410 } 411 case PathDiagnosticPiece::ControlFlow: 412 case PathDiagnosticPiece::Note: 413 case PathDiagnosticPiece::PopUp: 414 break; 415 } 416 path.push_back(std::move(piece)); 417 } 418 } 419 420 /// Recursively scan through a path and prune out calls and macros pieces 421 /// that aren't needed. Return true if afterwards the path contains 422 /// "interesting stuff" which means it shouldn't be pruned from the parent path. 423 static bool removeUnneededCalls(const PathDiagnosticConstruct &C, 424 PathPieces &pieces, 425 const PathSensitiveBugReport *R, 426 bool IsInteresting = false) { 427 bool containsSomethingInteresting = IsInteresting; 428 const unsigned N = pieces.size(); 429 430 for (unsigned i = 0 ; i < N ; ++i) { 431 // Remove the front piece from the path. If it is still something we 432 // want to keep once we are done, we will push it back on the end. 433 auto piece = std::move(pieces.front()); 434 pieces.pop_front(); 435 436 switch (piece->getKind()) { 437 case PathDiagnosticPiece::Call: { 438 auto &call = cast<PathDiagnosticCallPiece>(*piece); 439 // Check if the location context is interesting. 440 if (!removeUnneededCalls( 441 C, call.path, R, 442 R->isInteresting(C.getLocationContextFor(&call.path)))) 443 continue; 444 445 containsSomethingInteresting = true; 446 break; 447 } 448 case PathDiagnosticPiece::Macro: { 449 auto ¯o = cast<PathDiagnosticMacroPiece>(*piece); 450 if (!removeUnneededCalls(C, macro.subPieces, R, IsInteresting)) 451 continue; 452 containsSomethingInteresting = true; 453 break; 454 } 455 case PathDiagnosticPiece::Event: { 456 auto &event = cast<PathDiagnosticEventPiece>(*piece); 457 458 // We never throw away an event, but we do throw it away wholesale 459 // as part of a path if we throw the entire path away. 460 containsSomethingInteresting |= !event.isPrunable(); 461 break; 462 } 463 case PathDiagnosticPiece::ControlFlow: 464 case PathDiagnosticPiece::Note: 465 case PathDiagnosticPiece::PopUp: 466 break; 467 } 468 469 pieces.push_back(std::move(piece)); 470 } 471 472 return containsSomethingInteresting; 473 } 474 475 /// Same logic as above to remove extra pieces. 476 static void removePopUpNotes(PathPieces &Path) { 477 for (unsigned int i = 0; i < Path.size(); ++i) { 478 auto Piece = std::move(Path.front()); 479 Path.pop_front(); 480 if (!isa<PathDiagnosticPopUpPiece>(*Piece)) 481 Path.push_back(std::move(Piece)); 482 } 483 } 484 485 /// Returns true if the given decl has been implicitly given a body, either by 486 /// the analyzer or by the compiler proper. 487 static bool hasImplicitBody(const Decl *D) { 488 assert(D); 489 return D->isImplicit() || !D->hasBody(); 490 } 491 492 /// Recursively scan through a path and make sure that all call pieces have 493 /// valid locations. 494 static void 495 adjustCallLocations(PathPieces &Pieces, 496 PathDiagnosticLocation *LastCallLocation = nullptr) { 497 for (const auto &I : Pieces) { 498 auto *Call = dyn_cast<PathDiagnosticCallPiece>(I.get()); 499 500 if (!Call) 501 continue; 502 503 if (LastCallLocation) { 504 bool CallerIsImplicit = hasImplicitBody(Call->getCaller()); 505 if (CallerIsImplicit || !Call->callEnter.asLocation().isValid()) 506 Call->callEnter = *LastCallLocation; 507 if (CallerIsImplicit || !Call->callReturn.asLocation().isValid()) 508 Call->callReturn = *LastCallLocation; 509 } 510 511 // Recursively clean out the subclass. Keep this call around if 512 // it contains any informative diagnostics. 513 PathDiagnosticLocation *ThisCallLocation; 514 if (Call->callEnterWithin.asLocation().isValid() && 515 !hasImplicitBody(Call->getCallee())) 516 ThisCallLocation = &Call->callEnterWithin; 517 else 518 ThisCallLocation = &Call->callEnter; 519 520 assert(ThisCallLocation && "Outermost call has an invalid location"); 521 adjustCallLocations(Call->path, ThisCallLocation); 522 } 523 } 524 525 /// Remove edges in and out of C++ default initializer expressions. These are 526 /// for fields that have in-class initializers, as opposed to being initialized 527 /// explicitly in a constructor or braced list. 528 static void removeEdgesToDefaultInitializers(PathPieces &Pieces) { 529 for (PathPieces::iterator I = Pieces.begin(), E = Pieces.end(); I != E;) { 530 if (auto *C = dyn_cast<PathDiagnosticCallPiece>(I->get())) 531 removeEdgesToDefaultInitializers(C->path); 532 533 if (auto *M = dyn_cast<PathDiagnosticMacroPiece>(I->get())) 534 removeEdgesToDefaultInitializers(M->subPieces); 535 536 if (auto *CF = dyn_cast<PathDiagnosticControlFlowPiece>(I->get())) { 537 const Stmt *Start = CF->getStartLocation().asStmt(); 538 const Stmt *End = CF->getEndLocation().asStmt(); 539 if (isa_and_nonnull<CXXDefaultInitExpr>(Start)) { 540 I = Pieces.erase(I); 541 continue; 542 } else if (isa_and_nonnull<CXXDefaultInitExpr>(End)) { 543 PathPieces::iterator Next = std::next(I); 544 if (Next != E) { 545 if (auto *NextCF = 546 dyn_cast<PathDiagnosticControlFlowPiece>(Next->get())) { 547 NextCF->setStartLocation(CF->getStartLocation()); 548 } 549 } 550 I = Pieces.erase(I); 551 continue; 552 } 553 } 554 555 I++; 556 } 557 } 558 559 /// Remove all pieces with invalid locations as these cannot be serialized. 560 /// We might have pieces with invalid locations as a result of inlining Body 561 /// Farm generated functions. 562 static void removePiecesWithInvalidLocations(PathPieces &Pieces) { 563 for (PathPieces::iterator I = Pieces.begin(), E = Pieces.end(); I != E;) { 564 if (auto *C = dyn_cast<PathDiagnosticCallPiece>(I->get())) 565 removePiecesWithInvalidLocations(C->path); 566 567 if (auto *M = dyn_cast<PathDiagnosticMacroPiece>(I->get())) 568 removePiecesWithInvalidLocations(M->subPieces); 569 570 if (!(*I)->getLocation().isValid() || 571 !(*I)->getLocation().asLocation().isValid()) { 572 I = Pieces.erase(I); 573 continue; 574 } 575 I++; 576 } 577 } 578 579 PathDiagnosticLocation PathDiagnosticBuilder::ExecutionContinues( 580 const PathDiagnosticConstruct &C) const { 581 if (const Stmt *S = C.getCurrentNode()->getNextStmtForDiagnostics()) 582 return PathDiagnosticLocation(S, getSourceManager(), 583 C.getCurrLocationContext()); 584 585 return PathDiagnosticLocation::createDeclEnd(C.getCurrLocationContext(), 586 getSourceManager()); 587 } 588 589 PathDiagnosticLocation PathDiagnosticBuilder::ExecutionContinues( 590 llvm::raw_string_ostream &os, const PathDiagnosticConstruct &C) const { 591 // Slow, but probably doesn't matter. 592 if (os.str().empty()) 593 os << ' '; 594 595 const PathDiagnosticLocation &Loc = ExecutionContinues(C); 596 597 if (Loc.asStmt()) 598 os << "Execution continues on line " 599 << getSourceManager().getExpansionLineNumber(Loc.asLocation()) 600 << '.'; 601 else { 602 os << "Execution jumps to the end of the "; 603 const Decl *D = C.getCurrLocationContext()->getDecl(); 604 if (isa<ObjCMethodDecl>(D)) 605 os << "method"; 606 else if (isa<FunctionDecl>(D)) 607 os << "function"; 608 else { 609 assert(isa<BlockDecl>(D)); 610 os << "anonymous block"; 611 } 612 os << '.'; 613 } 614 615 return Loc; 616 } 617 618 static const Stmt *getEnclosingParent(const Stmt *S, const ParentMap &PM) { 619 if (isa<Expr>(S) && PM.isConsumedExpr(cast<Expr>(S))) 620 return PM.getParentIgnoreParens(S); 621 622 const Stmt *Parent = PM.getParentIgnoreParens(S); 623 if (!Parent) 624 return nullptr; 625 626 switch (Parent->getStmtClass()) { 627 case Stmt::ForStmtClass: 628 case Stmt::DoStmtClass: 629 case Stmt::WhileStmtClass: 630 case Stmt::ObjCForCollectionStmtClass: 631 case Stmt::CXXForRangeStmtClass: 632 return Parent; 633 default: 634 break; 635 } 636 637 return nullptr; 638 } 639 640 static PathDiagnosticLocation 641 getEnclosingStmtLocation(const Stmt *S, const LocationContext *LC, 642 bool allowNestedContexts = false) { 643 if (!S) 644 return {}; 645 646 const SourceManager &SMgr = LC->getDecl()->getASTContext().getSourceManager(); 647 648 while (const Stmt *Parent = getEnclosingParent(S, LC->getParentMap())) { 649 switch (Parent->getStmtClass()) { 650 case Stmt::BinaryOperatorClass: { 651 const auto *B = cast<BinaryOperator>(Parent); 652 if (B->isLogicalOp()) 653 return PathDiagnosticLocation(allowNestedContexts ? B : S, SMgr, LC); 654 break; 655 } 656 case Stmt::CompoundStmtClass: 657 case Stmt::StmtExprClass: 658 return PathDiagnosticLocation(S, SMgr, LC); 659 case Stmt::ChooseExprClass: 660 // Similar to '?' if we are referring to condition, just have the edge 661 // point to the entire choose expression. 662 if (allowNestedContexts || cast<ChooseExpr>(Parent)->getCond() == S) 663 return PathDiagnosticLocation(Parent, SMgr, LC); 664 else 665 return PathDiagnosticLocation(S, SMgr, LC); 666 case Stmt::BinaryConditionalOperatorClass: 667 case Stmt::ConditionalOperatorClass: 668 // For '?', if we are referring to condition, just have the edge point 669 // to the entire '?' expression. 670 if (allowNestedContexts || 671 cast<AbstractConditionalOperator>(Parent)->getCond() == S) 672 return PathDiagnosticLocation(Parent, SMgr, LC); 673 else 674 return PathDiagnosticLocation(S, SMgr, LC); 675 case Stmt::CXXForRangeStmtClass: 676 if (cast<CXXForRangeStmt>(Parent)->getBody() == S) 677 return PathDiagnosticLocation(S, SMgr, LC); 678 break; 679 case Stmt::DoStmtClass: 680 return PathDiagnosticLocation(S, SMgr, LC); 681 case Stmt::ForStmtClass: 682 if (cast<ForStmt>(Parent)->getBody() == S) 683 return PathDiagnosticLocation(S, SMgr, LC); 684 break; 685 case Stmt::IfStmtClass: 686 if (cast<IfStmt>(Parent)->getCond() != S) 687 return PathDiagnosticLocation(S, SMgr, LC); 688 break; 689 case Stmt::ObjCForCollectionStmtClass: 690 if (cast<ObjCForCollectionStmt>(Parent)->getBody() == S) 691 return PathDiagnosticLocation(S, SMgr, LC); 692 break; 693 case Stmt::WhileStmtClass: 694 if (cast<WhileStmt>(Parent)->getCond() != S) 695 return PathDiagnosticLocation(S, SMgr, LC); 696 break; 697 default: 698 break; 699 } 700 701 S = Parent; 702 } 703 704 assert(S && "Cannot have null Stmt for PathDiagnosticLocation"); 705 706 return PathDiagnosticLocation(S, SMgr, LC); 707 } 708 709 //===----------------------------------------------------------------------===// 710 // "Minimal" path diagnostic generation algorithm. 711 //===----------------------------------------------------------------------===// 712 713 /// If the piece contains a special message, add it to all the call pieces on 714 /// the active stack. For example, my_malloc allocated memory, so MallocChecker 715 /// will construct an event at the call to malloc(), and add a stack hint that 716 /// an allocated memory was returned. We'll use this hint to construct a message 717 /// when returning from the call to my_malloc 718 /// 719 /// void *my_malloc() { return malloc(sizeof(int)); } 720 /// void fishy() { 721 /// void *ptr = my_malloc(); // returned allocated memory 722 /// } // leak 723 void PathDiagnosticBuilder::updateStackPiecesWithMessage( 724 PathDiagnosticPieceRef P, const CallWithEntryStack &CallStack) const { 725 if (R->hasCallStackHint(P)) 726 for (const auto &I : CallStack) { 727 PathDiagnosticCallPiece *CP = I.first; 728 const ExplodedNode *N = I.second; 729 std::string stackMsg = R->getCallStackMessage(P, N); 730 731 // The last message on the path to final bug is the most important 732 // one. Since we traverse the path backwards, do not add the message 733 // if one has been previously added. 734 if (!CP->hasCallStackMessage()) 735 CP->setCallStackMessage(stackMsg); 736 } 737 } 738 739 static void CompactMacroExpandedPieces(PathPieces &path, 740 const SourceManager& SM); 741 742 PathDiagnosticPieceRef PathDiagnosticBuilder::generateDiagForSwitchOP( 743 const PathDiagnosticConstruct &C, const CFGBlock *Dst, 744 PathDiagnosticLocation &Start) const { 745 746 const SourceManager &SM = getSourceManager(); 747 // Figure out what case arm we took. 748 std::string sbuf; 749 llvm::raw_string_ostream os(sbuf); 750 PathDiagnosticLocation End; 751 752 if (const Stmt *S = Dst->getLabel()) { 753 End = PathDiagnosticLocation(S, SM, C.getCurrLocationContext()); 754 755 switch (S->getStmtClass()) { 756 default: 757 os << "No cases match in the switch statement. " 758 "Control jumps to line " 759 << End.asLocation().getExpansionLineNumber(); 760 break; 761 case Stmt::DefaultStmtClass: 762 os << "Control jumps to the 'default' case at line " 763 << End.asLocation().getExpansionLineNumber(); 764 break; 765 766 case Stmt::CaseStmtClass: { 767 os << "Control jumps to 'case "; 768 const auto *Case = cast<CaseStmt>(S); 769 const Expr *LHS = Case->getLHS()->IgnoreParenImpCasts(); 770 771 // Determine if it is an enum. 772 bool GetRawInt = true; 773 774 if (const auto *DR = dyn_cast<DeclRefExpr>(LHS)) { 775 // FIXME: Maybe this should be an assertion. Are there cases 776 // were it is not an EnumConstantDecl? 777 const auto *D = dyn_cast<EnumConstantDecl>(DR->getDecl()); 778 779 if (D) { 780 GetRawInt = false; 781 os << *D; 782 } 783 } 784 785 if (GetRawInt) 786 os << LHS->EvaluateKnownConstInt(getASTContext()); 787 788 os << ":' at line " << End.asLocation().getExpansionLineNumber(); 789 break; 790 } 791 } 792 } else { 793 os << "'Default' branch taken. "; 794 End = ExecutionContinues(os, C); 795 } 796 return std::make_shared<PathDiagnosticControlFlowPiece>(Start, End, 797 os.str()); 798 } 799 800 PathDiagnosticPieceRef PathDiagnosticBuilder::generateDiagForGotoOP( 801 const PathDiagnosticConstruct &C, const Stmt *S, 802 PathDiagnosticLocation &Start) const { 803 std::string sbuf; 804 llvm::raw_string_ostream os(sbuf); 805 const PathDiagnosticLocation &End = 806 getEnclosingStmtLocation(S, C.getCurrLocationContext()); 807 os << "Control jumps to line " << End.asLocation().getExpansionLineNumber(); 808 return std::make_shared<PathDiagnosticControlFlowPiece>(Start, End, os.str()); 809 } 810 811 PathDiagnosticPieceRef PathDiagnosticBuilder::generateDiagForBinaryOP( 812 const PathDiagnosticConstruct &C, const Stmt *T, const CFGBlock *Src, 813 const CFGBlock *Dst) const { 814 815 const SourceManager &SM = getSourceManager(); 816 817 const auto *B = cast<BinaryOperator>(T); 818 std::string sbuf; 819 llvm::raw_string_ostream os(sbuf); 820 os << "Left side of '"; 821 PathDiagnosticLocation Start, End; 822 823 if (B->getOpcode() == BO_LAnd) { 824 os << "&&" 825 << "' is "; 826 827 if (*(Src->succ_begin() + 1) == Dst) { 828 os << "false"; 829 End = PathDiagnosticLocation(B->getLHS(), SM, C.getCurrLocationContext()); 830 Start = 831 PathDiagnosticLocation::createOperatorLoc(B, SM); 832 } else { 833 os << "true"; 834 Start = 835 PathDiagnosticLocation(B->getLHS(), SM, C.getCurrLocationContext()); 836 End = ExecutionContinues(C); 837 } 838 } else { 839 assert(B->getOpcode() == BO_LOr); 840 os << "||" 841 << "' is "; 842 843 if (*(Src->succ_begin() + 1) == Dst) { 844 os << "false"; 845 Start = 846 PathDiagnosticLocation(B->getLHS(), SM, C.getCurrLocationContext()); 847 End = ExecutionContinues(C); 848 } else { 849 os << "true"; 850 End = PathDiagnosticLocation(B->getLHS(), SM, C.getCurrLocationContext()); 851 Start = 852 PathDiagnosticLocation::createOperatorLoc(B, SM); 853 } 854 } 855 return std::make_shared<PathDiagnosticControlFlowPiece>(Start, End, 856 os.str()); 857 } 858 859 void PathDiagnosticBuilder::generateMinimalDiagForBlockEdge( 860 PathDiagnosticConstruct &C, BlockEdge BE) const { 861 const SourceManager &SM = getSourceManager(); 862 const LocationContext *LC = C.getCurrLocationContext(); 863 const CFGBlock *Src = BE.getSrc(); 864 const CFGBlock *Dst = BE.getDst(); 865 const Stmt *T = Src->getTerminatorStmt(); 866 if (!T) 867 return; 868 869 auto Start = PathDiagnosticLocation::createBegin(T, SM, LC); 870 switch (T->getStmtClass()) { 871 default: 872 break; 873 874 case Stmt::GotoStmtClass: 875 case Stmt::IndirectGotoStmtClass: { 876 if (const Stmt *S = C.getCurrentNode()->getNextStmtForDiagnostics()) 877 C.getActivePath().push_front(generateDiagForGotoOP(C, S, Start)); 878 break; 879 } 880 881 case Stmt::SwitchStmtClass: { 882 C.getActivePath().push_front(generateDiagForSwitchOP(C, Dst, Start)); 883 break; 884 } 885 886 case Stmt::BreakStmtClass: 887 case Stmt::ContinueStmtClass: { 888 std::string sbuf; 889 llvm::raw_string_ostream os(sbuf); 890 PathDiagnosticLocation End = ExecutionContinues(os, C); 891 C.getActivePath().push_front( 892 std::make_shared<PathDiagnosticControlFlowPiece>(Start, End, os.str())); 893 break; 894 } 895 896 // Determine control-flow for ternary '?'. 897 case Stmt::BinaryConditionalOperatorClass: 898 case Stmt::ConditionalOperatorClass: { 899 std::string sbuf; 900 llvm::raw_string_ostream os(sbuf); 901 os << "'?' condition is "; 902 903 if (*(Src->succ_begin() + 1) == Dst) 904 os << "false"; 905 else 906 os << "true"; 907 908 PathDiagnosticLocation End = ExecutionContinues(C); 909 910 if (const Stmt *S = End.asStmt()) 911 End = getEnclosingStmtLocation(S, C.getCurrLocationContext()); 912 913 C.getActivePath().push_front( 914 std::make_shared<PathDiagnosticControlFlowPiece>(Start, End, os.str())); 915 break; 916 } 917 918 // Determine control-flow for short-circuited '&&' and '||'. 919 case Stmt::BinaryOperatorClass: { 920 if (!C.supportsLogicalOpControlFlow()) 921 break; 922 923 C.getActivePath().push_front(generateDiagForBinaryOP(C, T, Src, Dst)); 924 break; 925 } 926 927 case Stmt::DoStmtClass: 928 if (*(Src->succ_begin()) == Dst) { 929 std::string sbuf; 930 llvm::raw_string_ostream os(sbuf); 931 932 os << "Loop condition is true. "; 933 PathDiagnosticLocation End = ExecutionContinues(os, C); 934 935 if (const Stmt *S = End.asStmt()) 936 End = getEnclosingStmtLocation(S, C.getCurrLocationContext()); 937 938 C.getActivePath().push_front( 939 std::make_shared<PathDiagnosticControlFlowPiece>(Start, End, 940 os.str())); 941 } else { 942 PathDiagnosticLocation End = ExecutionContinues(C); 943 944 if (const Stmt *S = End.asStmt()) 945 End = getEnclosingStmtLocation(S, C.getCurrLocationContext()); 946 947 C.getActivePath().push_front( 948 std::make_shared<PathDiagnosticControlFlowPiece>( 949 Start, End, "Loop condition is false. Exiting loop")); 950 } 951 break; 952 953 case Stmt::WhileStmtClass: 954 case Stmt::ForStmtClass: 955 if (*(Src->succ_begin() + 1) == Dst) { 956 std::string sbuf; 957 llvm::raw_string_ostream os(sbuf); 958 959 os << "Loop condition is false. "; 960 PathDiagnosticLocation End = ExecutionContinues(os, C); 961 if (const Stmt *S = End.asStmt()) 962 End = getEnclosingStmtLocation(S, C.getCurrLocationContext()); 963 964 C.getActivePath().push_front( 965 std::make_shared<PathDiagnosticControlFlowPiece>(Start, End, 966 os.str())); 967 } else { 968 PathDiagnosticLocation End = ExecutionContinues(C); 969 if (const Stmt *S = End.asStmt()) 970 End = getEnclosingStmtLocation(S, C.getCurrLocationContext()); 971 972 C.getActivePath().push_front( 973 std::make_shared<PathDiagnosticControlFlowPiece>( 974 Start, End, "Loop condition is true. Entering loop body")); 975 } 976 977 break; 978 979 case Stmt::IfStmtClass: { 980 PathDiagnosticLocation End = ExecutionContinues(C); 981 982 if (const Stmt *S = End.asStmt()) 983 End = getEnclosingStmtLocation(S, C.getCurrLocationContext()); 984 985 if (*(Src->succ_begin() + 1) == Dst) 986 C.getActivePath().push_front( 987 std::make_shared<PathDiagnosticControlFlowPiece>( 988 Start, End, "Taking false branch")); 989 else 990 C.getActivePath().push_front( 991 std::make_shared<PathDiagnosticControlFlowPiece>( 992 Start, End, "Taking true branch")); 993 994 break; 995 } 996 } 997 } 998 999 //===----------------------------------------------------------------------===// 1000 // Functions for determining if a loop was executed 0 times. 1001 //===----------------------------------------------------------------------===// 1002 1003 static bool isLoop(const Stmt *Term) { 1004 switch (Term->getStmtClass()) { 1005 case Stmt::ForStmtClass: 1006 case Stmt::WhileStmtClass: 1007 case Stmt::ObjCForCollectionStmtClass: 1008 case Stmt::CXXForRangeStmtClass: 1009 return true; 1010 default: 1011 // Note that we intentionally do not include do..while here. 1012 return false; 1013 } 1014 } 1015 1016 static bool isJumpToFalseBranch(const BlockEdge *BE) { 1017 const CFGBlock *Src = BE->getSrc(); 1018 assert(Src->succ_size() == 2); 1019 return (*(Src->succ_begin()+1) == BE->getDst()); 1020 } 1021 1022 static bool isContainedByStmt(const ParentMap &PM, const Stmt *S, 1023 const Stmt *SubS) { 1024 while (SubS) { 1025 if (SubS == S) 1026 return true; 1027 SubS = PM.getParent(SubS); 1028 } 1029 return false; 1030 } 1031 1032 static const Stmt *getStmtBeforeCond(const ParentMap &PM, const Stmt *Term, 1033 const ExplodedNode *N) { 1034 while (N) { 1035 std::optional<StmtPoint> SP = N->getLocation().getAs<StmtPoint>(); 1036 if (SP) { 1037 const Stmt *S = SP->getStmt(); 1038 if (!isContainedByStmt(PM, Term, S)) 1039 return S; 1040 } 1041 N = N->getFirstPred(); 1042 } 1043 return nullptr; 1044 } 1045 1046 static bool isInLoopBody(const ParentMap &PM, const Stmt *S, const Stmt *Term) { 1047 const Stmt *LoopBody = nullptr; 1048 switch (Term->getStmtClass()) { 1049 case Stmt::CXXForRangeStmtClass: { 1050 const auto *FR = cast<CXXForRangeStmt>(Term); 1051 if (isContainedByStmt(PM, FR->getInc(), S)) 1052 return true; 1053 if (isContainedByStmt(PM, FR->getLoopVarStmt(), S)) 1054 return true; 1055 LoopBody = FR->getBody(); 1056 break; 1057 } 1058 case Stmt::ForStmtClass: { 1059 const auto *FS = cast<ForStmt>(Term); 1060 if (isContainedByStmt(PM, FS->getInc(), S)) 1061 return true; 1062 LoopBody = FS->getBody(); 1063 break; 1064 } 1065 case Stmt::ObjCForCollectionStmtClass: { 1066 const auto *FC = cast<ObjCForCollectionStmt>(Term); 1067 LoopBody = FC->getBody(); 1068 break; 1069 } 1070 case Stmt::WhileStmtClass: 1071 LoopBody = cast<WhileStmt>(Term)->getBody(); 1072 break; 1073 default: 1074 return false; 1075 } 1076 return isContainedByStmt(PM, LoopBody, S); 1077 } 1078 1079 /// Adds a sanitized control-flow diagnostic edge to a path. 1080 static void addEdgeToPath(PathPieces &path, 1081 PathDiagnosticLocation &PrevLoc, 1082 PathDiagnosticLocation NewLoc) { 1083 if (!NewLoc.isValid()) 1084 return; 1085 1086 SourceLocation NewLocL = NewLoc.asLocation(); 1087 if (NewLocL.isInvalid()) 1088 return; 1089 1090 if (!PrevLoc.isValid() || !PrevLoc.asLocation().isValid()) { 1091 PrevLoc = NewLoc; 1092 return; 1093 } 1094 1095 // Ignore self-edges, which occur when there are multiple nodes at the same 1096 // statement. 1097 if (NewLoc.asStmt() && NewLoc.asStmt() == PrevLoc.asStmt()) 1098 return; 1099 1100 path.push_front( 1101 std::make_shared<PathDiagnosticControlFlowPiece>(NewLoc, PrevLoc)); 1102 PrevLoc = NewLoc; 1103 } 1104 1105 /// A customized wrapper for CFGBlock::getTerminatorCondition() 1106 /// which returns the element for ObjCForCollectionStmts. 1107 static const Stmt *getTerminatorCondition(const CFGBlock *B) { 1108 const Stmt *S = B->getTerminatorCondition(); 1109 if (const auto *FS = dyn_cast_or_null<ObjCForCollectionStmt>(S)) 1110 return FS->getElement(); 1111 return S; 1112 } 1113 1114 constexpr llvm::StringLiteral StrEnteringLoop = "Entering loop body"; 1115 constexpr llvm::StringLiteral StrLoopBodyZero = "Loop body executed 0 times"; 1116 constexpr llvm::StringLiteral StrLoopRangeEmpty = 1117 "Loop body skipped when range is empty"; 1118 constexpr llvm::StringLiteral StrLoopCollectionEmpty = 1119 "Loop body skipped when collection is empty"; 1120 1121 static std::unique_ptr<FilesToLineNumsMap> 1122 findExecutedLines(const SourceManager &SM, const ExplodedNode *N); 1123 1124 void PathDiagnosticBuilder::generatePathDiagnosticsForNode( 1125 PathDiagnosticConstruct &C, PathDiagnosticLocation &PrevLoc) const { 1126 ProgramPoint P = C.getCurrentNode()->getLocation(); 1127 const SourceManager &SM = getSourceManager(); 1128 1129 // Have we encountered an entrance to a call? It may be 1130 // the case that we have not encountered a matching 1131 // call exit before this point. This means that the path 1132 // terminated within the call itself. 1133 if (auto CE = P.getAs<CallEnter>()) { 1134 1135 if (C.shouldAddPathEdges()) { 1136 // Add an edge to the start of the function. 1137 const StackFrameContext *CalleeLC = CE->getCalleeContext(); 1138 const Decl *D = CalleeLC->getDecl(); 1139 // Add the edge only when the callee has body. We jump to the beginning 1140 // of the *declaration*, however we expect it to be followed by the 1141 // body. This isn't the case for autosynthesized property accessors in 1142 // Objective-C. No need for a similar extra check for CallExit points 1143 // because the exit edge comes from a statement (i.e. return), 1144 // not from declaration. 1145 if (D->hasBody()) 1146 addEdgeToPath(C.getActivePath(), PrevLoc, 1147 PathDiagnosticLocation::createBegin(D, SM)); 1148 } 1149 1150 // Did we visit an entire call? 1151 bool VisitedEntireCall = C.PD->isWithinCall(); 1152 C.PD->popActivePath(); 1153 1154 PathDiagnosticCallPiece *Call; 1155 if (VisitedEntireCall) { 1156 Call = cast<PathDiagnosticCallPiece>(C.getActivePath().front().get()); 1157 } else { 1158 // The path terminated within a nested location context, create a new 1159 // call piece to encapsulate the rest of the path pieces. 1160 const Decl *Caller = CE->getLocationContext()->getDecl(); 1161 Call = PathDiagnosticCallPiece::construct(C.getActivePath(), Caller); 1162 assert(C.getActivePath().size() == 1 && 1163 C.getActivePath().front().get() == Call); 1164 1165 // Since we just transferred the path over to the call piece, reset the 1166 // mapping of the active path to the current location context. 1167 assert(C.isInLocCtxMap(&C.getActivePath()) && 1168 "When we ascend to a previously unvisited call, the active path's " 1169 "address shouldn't change, but rather should be compacted into " 1170 "a single CallEvent!"); 1171 C.updateLocCtxMap(&C.getActivePath(), C.getCurrLocationContext()); 1172 1173 // Record the location context mapping for the path within the call. 1174 assert(!C.isInLocCtxMap(&Call->path) && 1175 "When we ascend to a previously unvisited call, this must be the " 1176 "first time we encounter the caller context!"); 1177 C.updateLocCtxMap(&Call->path, CE->getCalleeContext()); 1178 } 1179 Call->setCallee(*CE, SM); 1180 1181 // Update the previous location in the active path. 1182 PrevLoc = Call->getLocation(); 1183 1184 if (!C.CallStack.empty()) { 1185 assert(C.CallStack.back().first == Call); 1186 C.CallStack.pop_back(); 1187 } 1188 return; 1189 } 1190 1191 assert(C.getCurrLocationContext() == C.getLocationContextForActivePath() && 1192 "The current position in the bug path is out of sync with the " 1193 "location context associated with the active path!"); 1194 1195 // Have we encountered an exit from a function call? 1196 if (std::optional<CallExitEnd> CE = P.getAs<CallExitEnd>()) { 1197 1198 // We are descending into a call (backwards). Construct 1199 // a new call piece to contain the path pieces for that call. 1200 auto Call = PathDiagnosticCallPiece::construct(*CE, SM); 1201 // Record the mapping from call piece to LocationContext. 1202 assert(!C.isInLocCtxMap(&Call->path) && 1203 "We just entered a call, this must've been the first time we " 1204 "encounter its context!"); 1205 C.updateLocCtxMap(&Call->path, CE->getCalleeContext()); 1206 1207 if (C.shouldAddPathEdges()) { 1208 // Add the edge to the return site. 1209 addEdgeToPath(C.getActivePath(), PrevLoc, Call->callReturn); 1210 PrevLoc.invalidate(); 1211 } 1212 1213 auto *P = Call.get(); 1214 C.getActivePath().push_front(std::move(Call)); 1215 1216 // Make the contents of the call the active path for now. 1217 C.PD->pushActivePath(&P->path); 1218 C.CallStack.push_back(CallWithEntry(P, C.getCurrentNode())); 1219 return; 1220 } 1221 1222 if (auto PS = P.getAs<PostStmt>()) { 1223 if (!C.shouldAddPathEdges()) 1224 return; 1225 1226 // Add an edge. If this is an ObjCForCollectionStmt do 1227 // not add an edge here as it appears in the CFG both 1228 // as a terminator and as a terminator condition. 1229 if (!isa<ObjCForCollectionStmt>(PS->getStmt())) { 1230 PathDiagnosticLocation L = 1231 PathDiagnosticLocation(PS->getStmt(), SM, C.getCurrLocationContext()); 1232 addEdgeToPath(C.getActivePath(), PrevLoc, L); 1233 } 1234 1235 } else if (auto BE = P.getAs<BlockEdge>()) { 1236 1237 if (C.shouldAddControlNotes()) { 1238 generateMinimalDiagForBlockEdge(C, *BE); 1239 } 1240 1241 if (!C.shouldAddPathEdges()) { 1242 return; 1243 } 1244 1245 // Are we jumping to the head of a loop? Add a special diagnostic. 1246 if (const Stmt *Loop = BE->getSrc()->getLoopTarget()) { 1247 PathDiagnosticLocation L(Loop, SM, C.getCurrLocationContext()); 1248 const Stmt *Body = nullptr; 1249 1250 if (const auto *FS = dyn_cast<ForStmt>(Loop)) 1251 Body = FS->getBody(); 1252 else if (const auto *WS = dyn_cast<WhileStmt>(Loop)) 1253 Body = WS->getBody(); 1254 else if (const auto *OFS = dyn_cast<ObjCForCollectionStmt>(Loop)) { 1255 Body = OFS->getBody(); 1256 } else if (const auto *FRS = dyn_cast<CXXForRangeStmt>(Loop)) { 1257 Body = FRS->getBody(); 1258 } 1259 // do-while statements are explicitly excluded here 1260 1261 auto p = std::make_shared<PathDiagnosticEventPiece>( 1262 L, "Looping back to the head of the loop"); 1263 p->setPrunable(true); 1264 1265 addEdgeToPath(C.getActivePath(), PrevLoc, p->getLocation()); 1266 // We might've added a very similar control node already 1267 if (!C.shouldAddControlNotes()) { 1268 C.getActivePath().push_front(std::move(p)); 1269 } 1270 1271 if (const auto *CS = dyn_cast_or_null<CompoundStmt>(Body)) { 1272 addEdgeToPath(C.getActivePath(), PrevLoc, 1273 PathDiagnosticLocation::createEndBrace(CS, SM)); 1274 } 1275 } 1276 1277 const CFGBlock *BSrc = BE->getSrc(); 1278 const ParentMap &PM = C.getParentMap(); 1279 1280 if (const Stmt *Term = BSrc->getTerminatorStmt()) { 1281 // Are we jumping past the loop body without ever executing the 1282 // loop (because the condition was false)? 1283 if (isLoop(Term)) { 1284 const Stmt *TermCond = getTerminatorCondition(BSrc); 1285 bool IsInLoopBody = isInLoopBody( 1286 PM, getStmtBeforeCond(PM, TermCond, C.getCurrentNode()), Term); 1287 1288 StringRef str; 1289 1290 if (isJumpToFalseBranch(&*BE)) { 1291 if (!IsInLoopBody) { 1292 if (isa<ObjCForCollectionStmt>(Term)) { 1293 str = StrLoopCollectionEmpty; 1294 } else if (isa<CXXForRangeStmt>(Term)) { 1295 str = StrLoopRangeEmpty; 1296 } else { 1297 str = StrLoopBodyZero; 1298 } 1299 } 1300 } else { 1301 str = StrEnteringLoop; 1302 } 1303 1304 if (!str.empty()) { 1305 PathDiagnosticLocation L(TermCond ? TermCond : Term, SM, 1306 C.getCurrLocationContext()); 1307 auto PE = std::make_shared<PathDiagnosticEventPiece>(L, str); 1308 PE->setPrunable(true); 1309 addEdgeToPath(C.getActivePath(), PrevLoc, PE->getLocation()); 1310 1311 // We might've added a very similar control node already 1312 if (!C.shouldAddControlNotes()) { 1313 C.getActivePath().push_front(std::move(PE)); 1314 } 1315 } 1316 } else if (isa<BreakStmt, ContinueStmt, GotoStmt>(Term)) { 1317 PathDiagnosticLocation L(Term, SM, C.getCurrLocationContext()); 1318 addEdgeToPath(C.getActivePath(), PrevLoc, L); 1319 } 1320 } 1321 } 1322 } 1323 1324 static std::unique_ptr<PathDiagnostic> 1325 generateDiagnosticForBasicReport(const BasicBugReport *R) { 1326 const BugType &BT = R->getBugType(); 1327 return std::make_unique<PathDiagnostic>( 1328 BT.getCheckerName(), R->getDeclWithIssue(), BT.getDescription(), 1329 R->getDescription(), R->getShortDescription(/*UseFallback=*/false), 1330 BT.getCategory(), R->getUniqueingLocation(), R->getUniqueingDecl(), 1331 std::make_unique<FilesToLineNumsMap>()); 1332 } 1333 1334 static std::unique_ptr<PathDiagnostic> 1335 generateEmptyDiagnosticForReport(const PathSensitiveBugReport *R, 1336 const SourceManager &SM) { 1337 const BugType &BT = R->getBugType(); 1338 return std::make_unique<PathDiagnostic>( 1339 BT.getCheckerName(), R->getDeclWithIssue(), BT.getDescription(), 1340 R->getDescription(), R->getShortDescription(/*UseFallback=*/false), 1341 BT.getCategory(), R->getUniqueingLocation(), R->getUniqueingDecl(), 1342 findExecutedLines(SM, R->getErrorNode())); 1343 } 1344 1345 static const Stmt *getStmtParent(const Stmt *S, const ParentMap &PM) { 1346 if (!S) 1347 return nullptr; 1348 1349 while (true) { 1350 S = PM.getParentIgnoreParens(S); 1351 1352 if (!S) 1353 break; 1354 1355 if (isa<FullExpr, CXXBindTemporaryExpr, SubstNonTypeTemplateParmExpr>(S)) 1356 continue; 1357 1358 break; 1359 } 1360 1361 return S; 1362 } 1363 1364 static bool isConditionForTerminator(const Stmt *S, const Stmt *Cond) { 1365 switch (S->getStmtClass()) { 1366 case Stmt::BinaryOperatorClass: { 1367 const auto *BO = cast<BinaryOperator>(S); 1368 if (!BO->isLogicalOp()) 1369 return false; 1370 return BO->getLHS() == Cond || BO->getRHS() == Cond; 1371 } 1372 case Stmt::IfStmtClass: 1373 return cast<IfStmt>(S)->getCond() == Cond; 1374 case Stmt::ForStmtClass: 1375 return cast<ForStmt>(S)->getCond() == Cond; 1376 case Stmt::WhileStmtClass: 1377 return cast<WhileStmt>(S)->getCond() == Cond; 1378 case Stmt::DoStmtClass: 1379 return cast<DoStmt>(S)->getCond() == Cond; 1380 case Stmt::ChooseExprClass: 1381 return cast<ChooseExpr>(S)->getCond() == Cond; 1382 case Stmt::IndirectGotoStmtClass: 1383 return cast<IndirectGotoStmt>(S)->getTarget() == Cond; 1384 case Stmt::SwitchStmtClass: 1385 return cast<SwitchStmt>(S)->getCond() == Cond; 1386 case Stmt::BinaryConditionalOperatorClass: 1387 return cast<BinaryConditionalOperator>(S)->getCond() == Cond; 1388 case Stmt::ConditionalOperatorClass: { 1389 const auto *CO = cast<ConditionalOperator>(S); 1390 return CO->getCond() == Cond || 1391 CO->getLHS() == Cond || 1392 CO->getRHS() == Cond; 1393 } 1394 case Stmt::ObjCForCollectionStmtClass: 1395 return cast<ObjCForCollectionStmt>(S)->getElement() == Cond; 1396 case Stmt::CXXForRangeStmtClass: { 1397 const auto *FRS = cast<CXXForRangeStmt>(S); 1398 return FRS->getCond() == Cond || FRS->getRangeInit() == Cond; 1399 } 1400 default: 1401 return false; 1402 } 1403 } 1404 1405 static bool isIncrementOrInitInForLoop(const Stmt *S, const Stmt *FL) { 1406 if (const auto *FS = dyn_cast<ForStmt>(FL)) 1407 return FS->getInc() == S || FS->getInit() == S; 1408 if (const auto *FRS = dyn_cast<CXXForRangeStmt>(FL)) 1409 return FRS->getInc() == S || FRS->getRangeStmt() == S || 1410 FRS->getLoopVarStmt() || FRS->getRangeInit() == S; 1411 return false; 1412 } 1413 1414 using OptimizedCallsSet = llvm::DenseSet<const PathDiagnosticCallPiece *>; 1415 1416 /// Adds synthetic edges from top-level statements to their subexpressions. 1417 /// 1418 /// This avoids a "swoosh" effect, where an edge from a top-level statement A 1419 /// points to a sub-expression B.1 that's not at the start of B. In these cases, 1420 /// we'd like to see an edge from A to B, then another one from B to B.1. 1421 static void addContextEdges(PathPieces &pieces, const LocationContext *LC) { 1422 const ParentMap &PM = LC->getParentMap(); 1423 PathPieces::iterator Prev = pieces.end(); 1424 for (PathPieces::iterator I = pieces.begin(), E = Prev; I != E; 1425 Prev = I, ++I) { 1426 auto *Piece = dyn_cast<PathDiagnosticControlFlowPiece>(I->get()); 1427 1428 if (!Piece) 1429 continue; 1430 1431 PathDiagnosticLocation SrcLoc = Piece->getStartLocation(); 1432 SmallVector<PathDiagnosticLocation, 4> SrcContexts; 1433 1434 PathDiagnosticLocation NextSrcContext = SrcLoc; 1435 const Stmt *InnerStmt = nullptr; 1436 while (NextSrcContext.isValid() && NextSrcContext.asStmt() != InnerStmt) { 1437 SrcContexts.push_back(NextSrcContext); 1438 InnerStmt = NextSrcContext.asStmt(); 1439 NextSrcContext = getEnclosingStmtLocation(InnerStmt, LC, 1440 /*allowNested=*/true); 1441 } 1442 1443 // Repeatedly split the edge as necessary. 1444 // This is important for nested logical expressions (||, &&, ?:) where we 1445 // want to show all the levels of context. 1446 while (true) { 1447 const Stmt *Dst = Piece->getEndLocation().getStmtOrNull(); 1448 1449 // We are looking at an edge. Is the destination within a larger 1450 // expression? 1451 PathDiagnosticLocation DstContext = 1452 getEnclosingStmtLocation(Dst, LC, /*allowNested=*/true); 1453 if (!DstContext.isValid() || DstContext.asStmt() == Dst) 1454 break; 1455 1456 // If the source is in the same context, we're already good. 1457 if (llvm::is_contained(SrcContexts, DstContext)) 1458 break; 1459 1460 // Update the subexpression node to point to the context edge. 1461 Piece->setStartLocation(DstContext); 1462 1463 // Try to extend the previous edge if it's at the same level as the source 1464 // context. 1465 if (Prev != E) { 1466 auto *PrevPiece = dyn_cast<PathDiagnosticControlFlowPiece>(Prev->get()); 1467 1468 if (PrevPiece) { 1469 if (const Stmt *PrevSrc = 1470 PrevPiece->getStartLocation().getStmtOrNull()) { 1471 const Stmt *PrevSrcParent = getStmtParent(PrevSrc, PM); 1472 if (PrevSrcParent == 1473 getStmtParent(DstContext.getStmtOrNull(), PM)) { 1474 PrevPiece->setEndLocation(DstContext); 1475 break; 1476 } 1477 } 1478 } 1479 } 1480 1481 // Otherwise, split the current edge into a context edge and a 1482 // subexpression edge. Note that the context statement may itself have 1483 // context. 1484 auto P = 1485 std::make_shared<PathDiagnosticControlFlowPiece>(SrcLoc, DstContext); 1486 Piece = P.get(); 1487 I = pieces.insert(I, std::move(P)); 1488 } 1489 } 1490 } 1491 1492 /// Move edges from a branch condition to a branch target 1493 /// when the condition is simple. 1494 /// 1495 /// This restructures some of the work of addContextEdges. That function 1496 /// creates edges this may destroy, but they work together to create a more 1497 /// aesthetically set of edges around branches. After the call to 1498 /// addContextEdges, we may have (1) an edge to the branch, (2) an edge from 1499 /// the branch to the branch condition, and (3) an edge from the branch 1500 /// condition to the branch target. We keep (1), but may wish to remove (2) 1501 /// and move the source of (3) to the branch if the branch condition is simple. 1502 static void simplifySimpleBranches(PathPieces &pieces) { 1503 for (PathPieces::iterator I = pieces.begin(), E = pieces.end(); I != E; ++I) { 1504 const auto *PieceI = dyn_cast<PathDiagnosticControlFlowPiece>(I->get()); 1505 1506 if (!PieceI) 1507 continue; 1508 1509 const Stmt *s1Start = PieceI->getStartLocation().getStmtOrNull(); 1510 const Stmt *s1End = PieceI->getEndLocation().getStmtOrNull(); 1511 1512 if (!s1Start || !s1End) 1513 continue; 1514 1515 PathPieces::iterator NextI = I; ++NextI; 1516 if (NextI == E) 1517 break; 1518 1519 PathDiagnosticControlFlowPiece *PieceNextI = nullptr; 1520 1521 while (true) { 1522 if (NextI == E) 1523 break; 1524 1525 const auto *EV = dyn_cast<PathDiagnosticEventPiece>(NextI->get()); 1526 if (EV) { 1527 StringRef S = EV->getString(); 1528 if (S == StrEnteringLoop || S == StrLoopBodyZero || 1529 S == StrLoopCollectionEmpty || S == StrLoopRangeEmpty) { 1530 ++NextI; 1531 continue; 1532 } 1533 break; 1534 } 1535 1536 PieceNextI = dyn_cast<PathDiagnosticControlFlowPiece>(NextI->get()); 1537 break; 1538 } 1539 1540 if (!PieceNextI) 1541 continue; 1542 1543 const Stmt *s2Start = PieceNextI->getStartLocation().getStmtOrNull(); 1544 const Stmt *s2End = PieceNextI->getEndLocation().getStmtOrNull(); 1545 1546 if (!s2Start || !s2End || s1End != s2Start) 1547 continue; 1548 1549 // We only perform this transformation for specific branch kinds. 1550 // We don't want to do this for do..while, for example. 1551 if (!isa<ForStmt, WhileStmt, IfStmt, ObjCForCollectionStmt, 1552 CXXForRangeStmt>(s1Start)) 1553 continue; 1554 1555 // Is s1End the branch condition? 1556 if (!isConditionForTerminator(s1Start, s1End)) 1557 continue; 1558 1559 // Perform the hoisting by eliminating (2) and changing the start 1560 // location of (3). 1561 PieceNextI->setStartLocation(PieceI->getStartLocation()); 1562 I = pieces.erase(I); 1563 } 1564 } 1565 1566 /// Returns the number of bytes in the given (character-based) SourceRange. 1567 /// 1568 /// If the locations in the range are not on the same line, returns 1569 /// std::nullopt. 1570 /// 1571 /// Note that this does not do a precise user-visible character or column count. 1572 static std::optional<size_t> getLengthOnSingleLine(const SourceManager &SM, 1573 SourceRange Range) { 1574 SourceRange ExpansionRange(SM.getExpansionLoc(Range.getBegin()), 1575 SM.getExpansionRange(Range.getEnd()).getEnd()); 1576 1577 FileID FID = SM.getFileID(ExpansionRange.getBegin()); 1578 if (FID != SM.getFileID(ExpansionRange.getEnd())) 1579 return std::nullopt; 1580 1581 std::optional<MemoryBufferRef> Buffer = SM.getBufferOrNone(FID); 1582 if (!Buffer) 1583 return std::nullopt; 1584 1585 unsigned BeginOffset = SM.getFileOffset(ExpansionRange.getBegin()); 1586 unsigned EndOffset = SM.getFileOffset(ExpansionRange.getEnd()); 1587 StringRef Snippet = Buffer->getBuffer().slice(BeginOffset, EndOffset); 1588 1589 // We're searching the raw bytes of the buffer here, which might include 1590 // escaped newlines and such. That's okay; we're trying to decide whether the 1591 // SourceRange is covering a large or small amount of space in the user's 1592 // editor. 1593 if (Snippet.find_first_of("\r\n") != StringRef::npos) 1594 return std::nullopt; 1595 1596 // This isn't Unicode-aware, but it doesn't need to be. 1597 return Snippet.size(); 1598 } 1599 1600 /// \sa getLengthOnSingleLine(SourceManager, SourceRange) 1601 static std::optional<size_t> getLengthOnSingleLine(const SourceManager &SM, 1602 const Stmt *S) { 1603 return getLengthOnSingleLine(SM, S->getSourceRange()); 1604 } 1605 1606 /// Eliminate two-edge cycles created by addContextEdges(). 1607 /// 1608 /// Once all the context edges are in place, there are plenty of cases where 1609 /// there's a single edge from a top-level statement to a subexpression, 1610 /// followed by a single path note, and then a reverse edge to get back out to 1611 /// the top level. If the statement is simple enough, the subexpression edges 1612 /// just add noise and make it harder to understand what's going on. 1613 /// 1614 /// This function only removes edges in pairs, because removing only one edge 1615 /// might leave other edges dangling. 1616 /// 1617 /// This will not remove edges in more complicated situations: 1618 /// - if there is more than one "hop" leading to or from a subexpression. 1619 /// - if there is an inlined call between the edges instead of a single event. 1620 /// - if the whole statement is large enough that having subexpression arrows 1621 /// might be helpful. 1622 static void removeContextCycles(PathPieces &Path, const SourceManager &SM) { 1623 for (PathPieces::iterator I = Path.begin(), E = Path.end(); I != E; ) { 1624 // Pattern match the current piece and its successor. 1625 const auto *PieceI = dyn_cast<PathDiagnosticControlFlowPiece>(I->get()); 1626 1627 if (!PieceI) { 1628 ++I; 1629 continue; 1630 } 1631 1632 const Stmt *s1Start = PieceI->getStartLocation().getStmtOrNull(); 1633 const Stmt *s1End = PieceI->getEndLocation().getStmtOrNull(); 1634 1635 PathPieces::iterator NextI = I; ++NextI; 1636 if (NextI == E) 1637 break; 1638 1639 const auto *PieceNextI = 1640 dyn_cast<PathDiagnosticControlFlowPiece>(NextI->get()); 1641 1642 if (!PieceNextI) { 1643 if (isa<PathDiagnosticEventPiece>(NextI->get())) { 1644 ++NextI; 1645 if (NextI == E) 1646 break; 1647 PieceNextI = dyn_cast<PathDiagnosticControlFlowPiece>(NextI->get()); 1648 } 1649 1650 if (!PieceNextI) { 1651 ++I; 1652 continue; 1653 } 1654 } 1655 1656 const Stmt *s2Start = PieceNextI->getStartLocation().getStmtOrNull(); 1657 const Stmt *s2End = PieceNextI->getEndLocation().getStmtOrNull(); 1658 1659 if (s1Start && s2Start && s1Start == s2End && s2Start == s1End) { 1660 const size_t MAX_SHORT_LINE_LENGTH = 80; 1661 std::optional<size_t> s1Length = getLengthOnSingleLine(SM, s1Start); 1662 if (s1Length && *s1Length <= MAX_SHORT_LINE_LENGTH) { 1663 std::optional<size_t> s2Length = getLengthOnSingleLine(SM, s2Start); 1664 if (s2Length && *s2Length <= MAX_SHORT_LINE_LENGTH) { 1665 Path.erase(I); 1666 I = Path.erase(NextI); 1667 continue; 1668 } 1669 } 1670 } 1671 1672 ++I; 1673 } 1674 } 1675 1676 /// Return true if X is contained by Y. 1677 static bool lexicalContains(const ParentMap &PM, const Stmt *X, const Stmt *Y) { 1678 while (X) { 1679 if (X == Y) 1680 return true; 1681 X = PM.getParent(X); 1682 } 1683 return false; 1684 } 1685 1686 // Remove short edges on the same line less than 3 columns in difference. 1687 static void removePunyEdges(PathPieces &path, const SourceManager &SM, 1688 const ParentMap &PM) { 1689 bool erased = false; 1690 1691 for (PathPieces::iterator I = path.begin(), E = path.end(); I != E; 1692 erased ? I : ++I) { 1693 erased = false; 1694 1695 const auto *PieceI = dyn_cast<PathDiagnosticControlFlowPiece>(I->get()); 1696 1697 if (!PieceI) 1698 continue; 1699 1700 const Stmt *start = PieceI->getStartLocation().getStmtOrNull(); 1701 const Stmt *end = PieceI->getEndLocation().getStmtOrNull(); 1702 1703 if (!start || !end) 1704 continue; 1705 1706 const Stmt *endParent = PM.getParent(end); 1707 if (!endParent) 1708 continue; 1709 1710 if (isConditionForTerminator(end, endParent)) 1711 continue; 1712 1713 SourceLocation FirstLoc = start->getBeginLoc(); 1714 SourceLocation SecondLoc = end->getBeginLoc(); 1715 1716 if (!SM.isWrittenInSameFile(FirstLoc, SecondLoc)) 1717 continue; 1718 if (SM.isBeforeInTranslationUnit(SecondLoc, FirstLoc)) 1719 std::swap(SecondLoc, FirstLoc); 1720 1721 SourceRange EdgeRange(FirstLoc, SecondLoc); 1722 std::optional<size_t> ByteWidth = getLengthOnSingleLine(SM, EdgeRange); 1723 1724 // If the statements are on different lines, continue. 1725 if (!ByteWidth) 1726 continue; 1727 1728 const size_t MAX_PUNY_EDGE_LENGTH = 2; 1729 if (*ByteWidth <= MAX_PUNY_EDGE_LENGTH) { 1730 // FIXME: There are enough /bytes/ between the endpoints of the edge, but 1731 // there might not be enough /columns/. A proper user-visible column count 1732 // is probably too expensive, though. 1733 I = path.erase(I); 1734 erased = true; 1735 continue; 1736 } 1737 } 1738 } 1739 1740 static void removeIdenticalEvents(PathPieces &path) { 1741 for (PathPieces::iterator I = path.begin(), E = path.end(); I != E; ++I) { 1742 const auto *PieceI = dyn_cast<PathDiagnosticEventPiece>(I->get()); 1743 1744 if (!PieceI) 1745 continue; 1746 1747 PathPieces::iterator NextI = I; ++NextI; 1748 if (NextI == E) 1749 return; 1750 1751 const auto *PieceNextI = dyn_cast<PathDiagnosticEventPiece>(NextI->get()); 1752 1753 if (!PieceNextI) 1754 continue; 1755 1756 // Erase the second piece if it has the same exact message text. 1757 if (PieceI->getString() == PieceNextI->getString()) { 1758 path.erase(NextI); 1759 } 1760 } 1761 } 1762 1763 static bool optimizeEdges(const PathDiagnosticConstruct &C, PathPieces &path, 1764 OptimizedCallsSet &OCS) { 1765 bool hasChanges = false; 1766 const LocationContext *LC = C.getLocationContextFor(&path); 1767 assert(LC); 1768 const ParentMap &PM = LC->getParentMap(); 1769 const SourceManager &SM = C.getSourceManager(); 1770 1771 for (PathPieces::iterator I = path.begin(), E = path.end(); I != E; ) { 1772 // Optimize subpaths. 1773 if (auto *CallI = dyn_cast<PathDiagnosticCallPiece>(I->get())) { 1774 // Record the fact that a call has been optimized so we only do the 1775 // effort once. 1776 if (!OCS.count(CallI)) { 1777 while (optimizeEdges(C, CallI->path, OCS)) { 1778 } 1779 OCS.insert(CallI); 1780 } 1781 ++I; 1782 continue; 1783 } 1784 1785 // Pattern match the current piece and its successor. 1786 auto *PieceI = dyn_cast<PathDiagnosticControlFlowPiece>(I->get()); 1787 1788 if (!PieceI) { 1789 ++I; 1790 continue; 1791 } 1792 1793 const Stmt *s1Start = PieceI->getStartLocation().getStmtOrNull(); 1794 const Stmt *s1End = PieceI->getEndLocation().getStmtOrNull(); 1795 const Stmt *level1 = getStmtParent(s1Start, PM); 1796 const Stmt *level2 = getStmtParent(s1End, PM); 1797 1798 PathPieces::iterator NextI = I; ++NextI; 1799 if (NextI == E) 1800 break; 1801 1802 const auto *PieceNextI = dyn_cast<PathDiagnosticControlFlowPiece>(NextI->get()); 1803 1804 if (!PieceNextI) { 1805 ++I; 1806 continue; 1807 } 1808 1809 const Stmt *s2Start = PieceNextI->getStartLocation().getStmtOrNull(); 1810 const Stmt *s2End = PieceNextI->getEndLocation().getStmtOrNull(); 1811 const Stmt *level3 = getStmtParent(s2Start, PM); 1812 const Stmt *level4 = getStmtParent(s2End, PM); 1813 1814 // Rule I. 1815 // 1816 // If we have two consecutive control edges whose end/begin locations 1817 // are at the same level (e.g. statements or top-level expressions within 1818 // a compound statement, or siblings share a single ancestor expression), 1819 // then merge them if they have no interesting intermediate event. 1820 // 1821 // For example: 1822 // 1823 // (1.1 -> 1.2) -> (1.2 -> 1.3) becomes (1.1 -> 1.3) because the common 1824 // parent is '1'. Here 'x.y.z' represents the hierarchy of statements. 1825 // 1826 // NOTE: this will be limited later in cases where we add barriers 1827 // to prevent this optimization. 1828 if (level1 && level1 == level2 && level1 == level3 && level1 == level4) { 1829 PieceI->setEndLocation(PieceNextI->getEndLocation()); 1830 path.erase(NextI); 1831 hasChanges = true; 1832 continue; 1833 } 1834 1835 // Rule II. 1836 // 1837 // Eliminate edges between subexpressions and parent expressions 1838 // when the subexpression is consumed. 1839 // 1840 // NOTE: this will be limited later in cases where we add barriers 1841 // to prevent this optimization. 1842 if (s1End && s1End == s2Start && level2) { 1843 bool removeEdge = false; 1844 // Remove edges into the increment or initialization of a 1845 // loop that have no interleaving event. This means that 1846 // they aren't interesting. 1847 if (isIncrementOrInitInForLoop(s1End, level2)) 1848 removeEdge = true; 1849 // Next only consider edges that are not anchored on 1850 // the condition of a terminator. This are intermediate edges 1851 // that we might want to trim. 1852 else if (!isConditionForTerminator(level2, s1End)) { 1853 // Trim edges on expressions that are consumed by 1854 // the parent expression. 1855 if (isa<Expr>(s1End) && PM.isConsumedExpr(cast<Expr>(s1End))) { 1856 removeEdge = true; 1857 } 1858 // Trim edges where a lexical containment doesn't exist. 1859 // For example: 1860 // 1861 // X -> Y -> Z 1862 // 1863 // If 'Z' lexically contains Y (it is an ancestor) and 1864 // 'X' does not lexically contain Y (it is a descendant OR 1865 // it has no lexical relationship at all) then trim. 1866 // 1867 // This can eliminate edges where we dive into a subexpression 1868 // and then pop back out, etc. 1869 else if (s1Start && s2End && 1870 lexicalContains(PM, s2Start, s2End) && 1871 !lexicalContains(PM, s1End, s1Start)) { 1872 removeEdge = true; 1873 } 1874 // Trim edges from a subexpression back to the top level if the 1875 // subexpression is on a different line. 1876 // 1877 // A.1 -> A -> B 1878 // becomes 1879 // A.1 -> B 1880 // 1881 // These edges just look ugly and don't usually add anything. 1882 else if (s1Start && s2End && 1883 lexicalContains(PM, s1Start, s1End)) { 1884 SourceRange EdgeRange(PieceI->getEndLocation().asLocation(), 1885 PieceI->getStartLocation().asLocation()); 1886 if (!getLengthOnSingleLine(SM, EdgeRange)) 1887 removeEdge = true; 1888 } 1889 } 1890 1891 if (removeEdge) { 1892 PieceI->setEndLocation(PieceNextI->getEndLocation()); 1893 path.erase(NextI); 1894 hasChanges = true; 1895 continue; 1896 } 1897 } 1898 1899 // Optimize edges for ObjC fast-enumeration loops. 1900 // 1901 // (X -> collection) -> (collection -> element) 1902 // 1903 // becomes: 1904 // 1905 // (X -> element) 1906 if (s1End == s2Start) { 1907 const auto *FS = dyn_cast_or_null<ObjCForCollectionStmt>(level3); 1908 if (FS && FS->getCollection()->IgnoreParens() == s2Start && 1909 s2End == FS->getElement()) { 1910 PieceI->setEndLocation(PieceNextI->getEndLocation()); 1911 path.erase(NextI); 1912 hasChanges = true; 1913 continue; 1914 } 1915 } 1916 1917 // No changes at this index? Move to the next one. 1918 ++I; 1919 } 1920 1921 if (!hasChanges) { 1922 // Adjust edges into subexpressions to make them more uniform 1923 // and aesthetically pleasing. 1924 addContextEdges(path, LC); 1925 // Remove "cyclical" edges that include one or more context edges. 1926 removeContextCycles(path, SM); 1927 // Hoist edges originating from branch conditions to branches 1928 // for simple branches. 1929 simplifySimpleBranches(path); 1930 // Remove any puny edges left over after primary optimization pass. 1931 removePunyEdges(path, SM, PM); 1932 // Remove identical events. 1933 removeIdenticalEvents(path); 1934 } 1935 1936 return hasChanges; 1937 } 1938 1939 /// Drop the very first edge in a path, which should be a function entry edge. 1940 /// 1941 /// If the first edge is not a function entry edge (say, because the first 1942 /// statement had an invalid source location), this function does nothing. 1943 // FIXME: We should just generate invalid edges anyway and have the optimizer 1944 // deal with them. 1945 static void dropFunctionEntryEdge(const PathDiagnosticConstruct &C, 1946 PathPieces &Path) { 1947 const auto *FirstEdge = 1948 dyn_cast<PathDiagnosticControlFlowPiece>(Path.front().get()); 1949 if (!FirstEdge) 1950 return; 1951 1952 const Decl *D = C.getLocationContextFor(&Path)->getDecl(); 1953 PathDiagnosticLocation EntryLoc = 1954 PathDiagnosticLocation::createBegin(D, C.getSourceManager()); 1955 if (FirstEdge->getStartLocation() != EntryLoc) 1956 return; 1957 1958 Path.pop_front(); 1959 } 1960 1961 /// Populate executes lines with lines containing at least one diagnostics. 1962 static void updateExecutedLinesWithDiagnosticPieces(PathDiagnostic &PD) { 1963 1964 PathPieces path = PD.path.flatten(/*ShouldFlattenMacros=*/true); 1965 FilesToLineNumsMap &ExecutedLines = PD.getExecutedLines(); 1966 1967 for (const auto &P : path) { 1968 FullSourceLoc Loc = P->getLocation().asLocation().getExpansionLoc(); 1969 FileID FID = Loc.getFileID(); 1970 unsigned LineNo = Loc.getLineNumber(); 1971 assert(FID.isValid()); 1972 ExecutedLines[FID].insert(LineNo); 1973 } 1974 } 1975 1976 PathDiagnosticConstruct::PathDiagnosticConstruct( 1977 const PathDiagnosticConsumer *PDC, const ExplodedNode *ErrorNode, 1978 const PathSensitiveBugReport *R) 1979 : Consumer(PDC), CurrentNode(ErrorNode), 1980 SM(CurrentNode->getCodeDecl().getASTContext().getSourceManager()), 1981 PD(generateEmptyDiagnosticForReport(R, getSourceManager())) { 1982 LCM[&PD->getActivePath()] = ErrorNode->getLocationContext(); 1983 } 1984 1985 PathDiagnosticBuilder::PathDiagnosticBuilder( 1986 BugReporterContext BRC, std::unique_ptr<ExplodedGraph> BugPath, 1987 PathSensitiveBugReport *r, const ExplodedNode *ErrorNode, 1988 std::unique_ptr<VisitorsDiagnosticsTy> VisitorsDiagnostics) 1989 : BugReporterContext(BRC), BugPath(std::move(BugPath)), R(r), 1990 ErrorNode(ErrorNode), 1991 VisitorsDiagnostics(std::move(VisitorsDiagnostics)) {} 1992 1993 std::unique_ptr<PathDiagnostic> 1994 PathDiagnosticBuilder::generate(const PathDiagnosticConsumer *PDC) const { 1995 PathDiagnosticConstruct Construct(PDC, ErrorNode, R); 1996 1997 const SourceManager &SM = getSourceManager(); 1998 const AnalyzerOptions &Opts = getAnalyzerOptions(); 1999 2000 if (!PDC->shouldGenerateDiagnostics()) 2001 return generateEmptyDiagnosticForReport(R, getSourceManager()); 2002 2003 // Construct the final (warning) event for the bug report. 2004 auto EndNotes = VisitorsDiagnostics->find(ErrorNode); 2005 PathDiagnosticPieceRef LastPiece; 2006 if (EndNotes != VisitorsDiagnostics->end()) { 2007 assert(!EndNotes->second.empty()); 2008 LastPiece = EndNotes->second[0]; 2009 } else { 2010 LastPiece = BugReporterVisitor::getDefaultEndPath(*this, ErrorNode, 2011 *getBugReport()); 2012 } 2013 Construct.PD->setEndOfPath(LastPiece); 2014 2015 PathDiagnosticLocation PrevLoc = Construct.PD->getLocation(); 2016 // From the error node to the root, ascend the bug path and construct the bug 2017 // report. 2018 while (Construct.ascendToPrevNode()) { 2019 generatePathDiagnosticsForNode(Construct, PrevLoc); 2020 2021 auto VisitorNotes = VisitorsDiagnostics->find(Construct.getCurrentNode()); 2022 if (VisitorNotes == VisitorsDiagnostics->end()) 2023 continue; 2024 2025 // This is a workaround due to inability to put shared PathDiagnosticPiece 2026 // into a FoldingSet. 2027 std::set<llvm::FoldingSetNodeID> DeduplicationSet; 2028 2029 // Add pieces from custom visitors. 2030 for (const PathDiagnosticPieceRef &Note : VisitorNotes->second) { 2031 llvm::FoldingSetNodeID ID; 2032 Note->Profile(ID); 2033 if (!DeduplicationSet.insert(ID).second) 2034 continue; 2035 2036 if (PDC->shouldAddPathEdges()) 2037 addEdgeToPath(Construct.getActivePath(), PrevLoc, Note->getLocation()); 2038 updateStackPiecesWithMessage(Note, Construct.CallStack); 2039 Construct.getActivePath().push_front(Note); 2040 } 2041 } 2042 2043 if (PDC->shouldAddPathEdges()) { 2044 // Add an edge to the start of the function. 2045 // We'll prune it out later, but it helps make diagnostics more uniform. 2046 const StackFrameContext *CalleeLC = 2047 Construct.getLocationContextForActivePath()->getStackFrame(); 2048 const Decl *D = CalleeLC->getDecl(); 2049 addEdgeToPath(Construct.getActivePath(), PrevLoc, 2050 PathDiagnosticLocation::createBegin(D, SM)); 2051 } 2052 2053 2054 // Finally, prune the diagnostic path of uninteresting stuff. 2055 if (!Construct.PD->path.empty()) { 2056 if (R->shouldPrunePath() && Opts.ShouldPrunePaths) { 2057 bool stillHasNotes = 2058 removeUnneededCalls(Construct, Construct.getMutablePieces(), R); 2059 assert(stillHasNotes); 2060 (void)stillHasNotes; 2061 } 2062 2063 // Remove pop-up notes if needed. 2064 if (!Opts.ShouldAddPopUpNotes) 2065 removePopUpNotes(Construct.getMutablePieces()); 2066 2067 // Redirect all call pieces to have valid locations. 2068 adjustCallLocations(Construct.getMutablePieces()); 2069 removePiecesWithInvalidLocations(Construct.getMutablePieces()); 2070 2071 if (PDC->shouldAddPathEdges()) { 2072 2073 // Reduce the number of edges from a very conservative set 2074 // to an aesthetically pleasing subset that conveys the 2075 // necessary information. 2076 OptimizedCallsSet OCS; 2077 while (optimizeEdges(Construct, Construct.getMutablePieces(), OCS)) { 2078 } 2079 2080 // Drop the very first function-entry edge. It's not really necessary 2081 // for top-level functions. 2082 dropFunctionEntryEdge(Construct, Construct.getMutablePieces()); 2083 } 2084 2085 // Remove messages that are basically the same, and edges that may not 2086 // make sense. 2087 // We have to do this after edge optimization in the Extensive mode. 2088 removeRedundantMsgs(Construct.getMutablePieces()); 2089 removeEdgesToDefaultInitializers(Construct.getMutablePieces()); 2090 } 2091 2092 if (Opts.ShouldDisplayMacroExpansions) 2093 CompactMacroExpandedPieces(Construct.getMutablePieces(), SM); 2094 2095 return std::move(Construct.PD); 2096 } 2097 2098 //===----------------------------------------------------------------------===// 2099 // Methods for BugType and subclasses. 2100 //===----------------------------------------------------------------------===// 2101 2102 void BugType::anchor() {} 2103 2104 void BuiltinBug::anchor() {} 2105 2106 //===----------------------------------------------------------------------===// 2107 // Methods for BugReport and subclasses. 2108 //===----------------------------------------------------------------------===// 2109 2110 LLVM_ATTRIBUTE_USED static bool 2111 isDependency(const CheckerRegistryData &Registry, StringRef CheckerName) { 2112 for (const std::pair<StringRef, StringRef> &Pair : Registry.Dependencies) { 2113 if (Pair.second == CheckerName) 2114 return true; 2115 } 2116 return false; 2117 } 2118 2119 LLVM_ATTRIBUTE_USED static bool isHidden(const CheckerRegistryData &Registry, 2120 StringRef CheckerName) { 2121 for (const CheckerInfo &Checker : Registry.Checkers) { 2122 if (Checker.FullName == CheckerName) 2123 return Checker.IsHidden; 2124 } 2125 llvm_unreachable( 2126 "Checker name not found in CheckerRegistry -- did you retrieve it " 2127 "correctly from CheckerManager::getCurrentCheckerName?"); 2128 } 2129 2130 PathSensitiveBugReport::PathSensitiveBugReport( 2131 const BugType &bt, StringRef shortDesc, StringRef desc, 2132 const ExplodedNode *errorNode, PathDiagnosticLocation LocationToUnique, 2133 const Decl *DeclToUnique) 2134 : BugReport(Kind::PathSensitive, bt, shortDesc, desc), ErrorNode(errorNode), 2135 ErrorNodeRange(getStmt() ? getStmt()->getSourceRange() : SourceRange()), 2136 UniqueingLocation(LocationToUnique), UniqueingDecl(DeclToUnique) { 2137 assert(!isDependency(ErrorNode->getState() 2138 ->getAnalysisManager() 2139 .getCheckerManager() 2140 ->getCheckerRegistryData(), 2141 bt.getCheckerName()) && 2142 "Some checkers depend on this one! We don't allow dependency " 2143 "checkers to emit warnings, because checkers should depend on " 2144 "*modeling*, not *diagnostics*."); 2145 2146 assert( 2147 (bt.getCheckerName().startswith("debug") || 2148 !isHidden(ErrorNode->getState() 2149 ->getAnalysisManager() 2150 .getCheckerManager() 2151 ->getCheckerRegistryData(), 2152 bt.getCheckerName())) && 2153 "Hidden checkers musn't emit diagnostics as they are by definition " 2154 "non-user facing!"); 2155 } 2156 2157 void PathSensitiveBugReport::addVisitor( 2158 std::unique_ptr<BugReporterVisitor> visitor) { 2159 if (!visitor) 2160 return; 2161 2162 llvm::FoldingSetNodeID ID; 2163 visitor->Profile(ID); 2164 2165 void *InsertPos = nullptr; 2166 if (CallbacksSet.FindNodeOrInsertPos(ID, InsertPos)) { 2167 return; 2168 } 2169 2170 Callbacks.push_back(std::move(visitor)); 2171 } 2172 2173 void PathSensitiveBugReport::clearVisitors() { 2174 Callbacks.clear(); 2175 } 2176 2177 const Decl *PathSensitiveBugReport::getDeclWithIssue() const { 2178 const ExplodedNode *N = getErrorNode(); 2179 if (!N) 2180 return nullptr; 2181 2182 const LocationContext *LC = N->getLocationContext(); 2183 return LC->getStackFrame()->getDecl(); 2184 } 2185 2186 void BasicBugReport::Profile(llvm::FoldingSetNodeID& hash) const { 2187 hash.AddInteger(static_cast<int>(getKind())); 2188 hash.AddPointer(&BT); 2189 hash.AddString(Description); 2190 assert(Location.isValid()); 2191 Location.Profile(hash); 2192 2193 for (SourceRange range : Ranges) { 2194 if (!range.isValid()) 2195 continue; 2196 hash.Add(range.getBegin()); 2197 hash.Add(range.getEnd()); 2198 } 2199 } 2200 2201 void PathSensitiveBugReport::Profile(llvm::FoldingSetNodeID &hash) const { 2202 hash.AddInteger(static_cast<int>(getKind())); 2203 hash.AddPointer(&BT); 2204 hash.AddString(Description); 2205 PathDiagnosticLocation UL = getUniqueingLocation(); 2206 if (UL.isValid()) { 2207 UL.Profile(hash); 2208 } else { 2209 // TODO: The statement may be null if the report was emitted before any 2210 // statements were executed. In particular, some checkers by design 2211 // occasionally emit their reports in empty functions (that have no 2212 // statements in their body). Do we profile correctly in this case? 2213 hash.AddPointer(ErrorNode->getCurrentOrPreviousStmtForDiagnostics()); 2214 } 2215 2216 for (SourceRange range : Ranges) { 2217 if (!range.isValid()) 2218 continue; 2219 hash.Add(range.getBegin()); 2220 hash.Add(range.getEnd()); 2221 } 2222 } 2223 2224 template <class T> 2225 static void insertToInterestingnessMap( 2226 llvm::DenseMap<T, bugreporter::TrackingKind> &InterestingnessMap, T Val, 2227 bugreporter::TrackingKind TKind) { 2228 auto Result = InterestingnessMap.insert({Val, TKind}); 2229 2230 if (Result.second) 2231 return; 2232 2233 // Even if this symbol/region was already marked as interesting as a 2234 // condition, if we later mark it as interesting again but with 2235 // thorough tracking, overwrite it. Entities marked with thorough 2236 // interestiness are the most important (or most interesting, if you will), 2237 // and we wouldn't like to downplay their importance. 2238 2239 switch (TKind) { 2240 case bugreporter::TrackingKind::Thorough: 2241 Result.first->getSecond() = bugreporter::TrackingKind::Thorough; 2242 return; 2243 case bugreporter::TrackingKind::Condition: 2244 return; 2245 } 2246 2247 llvm_unreachable( 2248 "BugReport::markInteresting currently can only handle 2 different " 2249 "tracking kinds! Please define what tracking kind should this entitiy" 2250 "have, if it was already marked as interesting with a different kind!"); 2251 } 2252 2253 void PathSensitiveBugReport::markInteresting(SymbolRef sym, 2254 bugreporter::TrackingKind TKind) { 2255 if (!sym) 2256 return; 2257 2258 insertToInterestingnessMap(InterestingSymbols, sym, TKind); 2259 2260 // FIXME: No tests exist for this code and it is questionable: 2261 // How to handle multiple metadata for the same region? 2262 if (const auto *meta = dyn_cast<SymbolMetadata>(sym)) 2263 markInteresting(meta->getRegion(), TKind); 2264 } 2265 2266 void PathSensitiveBugReport::markNotInteresting(SymbolRef sym) { 2267 if (!sym) 2268 return; 2269 InterestingSymbols.erase(sym); 2270 2271 // The metadata part of markInteresting is not reversed here. 2272 // Just making the same region not interesting is incorrect 2273 // in specific cases. 2274 if (const auto *meta = dyn_cast<SymbolMetadata>(sym)) 2275 markNotInteresting(meta->getRegion()); 2276 } 2277 2278 void PathSensitiveBugReport::markInteresting(const MemRegion *R, 2279 bugreporter::TrackingKind TKind) { 2280 if (!R) 2281 return; 2282 2283 R = R->getBaseRegion(); 2284 insertToInterestingnessMap(InterestingRegions, R, TKind); 2285 2286 if (const auto *SR = dyn_cast<SymbolicRegion>(R)) 2287 markInteresting(SR->getSymbol(), TKind); 2288 } 2289 2290 void PathSensitiveBugReport::markNotInteresting(const MemRegion *R) { 2291 if (!R) 2292 return; 2293 2294 R = R->getBaseRegion(); 2295 InterestingRegions.erase(R); 2296 2297 if (const auto *SR = dyn_cast<SymbolicRegion>(R)) 2298 markNotInteresting(SR->getSymbol()); 2299 } 2300 2301 void PathSensitiveBugReport::markInteresting(SVal V, 2302 bugreporter::TrackingKind TKind) { 2303 markInteresting(V.getAsRegion(), TKind); 2304 markInteresting(V.getAsSymbol(), TKind); 2305 } 2306 2307 void PathSensitiveBugReport::markInteresting(const LocationContext *LC) { 2308 if (!LC) 2309 return; 2310 InterestingLocationContexts.insert(LC); 2311 } 2312 2313 std::optional<bugreporter::TrackingKind> 2314 PathSensitiveBugReport::getInterestingnessKind(SVal V) const { 2315 auto RKind = getInterestingnessKind(V.getAsRegion()); 2316 auto SKind = getInterestingnessKind(V.getAsSymbol()); 2317 if (!RKind) 2318 return SKind; 2319 if (!SKind) 2320 return RKind; 2321 2322 // If either is marked with throrough tracking, return that, we wouldn't like 2323 // to downplay a note's importance by 'only' mentioning it as a condition. 2324 switch(*RKind) { 2325 case bugreporter::TrackingKind::Thorough: 2326 return RKind; 2327 case bugreporter::TrackingKind::Condition: 2328 return SKind; 2329 } 2330 2331 llvm_unreachable( 2332 "BugReport::getInterestingnessKind currently can only handle 2 different " 2333 "tracking kinds! Please define what tracking kind should we return here " 2334 "when the kind of getAsRegion() and getAsSymbol() is different!"); 2335 return std::nullopt; 2336 } 2337 2338 std::optional<bugreporter::TrackingKind> 2339 PathSensitiveBugReport::getInterestingnessKind(SymbolRef sym) const { 2340 if (!sym) 2341 return std::nullopt; 2342 // We don't currently consider metadata symbols to be interesting 2343 // even if we know their region is interesting. Is that correct behavior? 2344 auto It = InterestingSymbols.find(sym); 2345 if (It == InterestingSymbols.end()) 2346 return std::nullopt; 2347 return It->getSecond(); 2348 } 2349 2350 std::optional<bugreporter::TrackingKind> 2351 PathSensitiveBugReport::getInterestingnessKind(const MemRegion *R) const { 2352 if (!R) 2353 return std::nullopt; 2354 2355 R = R->getBaseRegion(); 2356 auto It = InterestingRegions.find(R); 2357 if (It != InterestingRegions.end()) 2358 return It->getSecond(); 2359 2360 if (const auto *SR = dyn_cast<SymbolicRegion>(R)) 2361 return getInterestingnessKind(SR->getSymbol()); 2362 return std::nullopt; 2363 } 2364 2365 bool PathSensitiveBugReport::isInteresting(SVal V) const { 2366 return getInterestingnessKind(V).has_value(); 2367 } 2368 2369 bool PathSensitiveBugReport::isInteresting(SymbolRef sym) const { 2370 return getInterestingnessKind(sym).has_value(); 2371 } 2372 2373 bool PathSensitiveBugReport::isInteresting(const MemRegion *R) const { 2374 return getInterestingnessKind(R).has_value(); 2375 } 2376 2377 bool PathSensitiveBugReport::isInteresting(const LocationContext *LC) const { 2378 if (!LC) 2379 return false; 2380 return InterestingLocationContexts.count(LC); 2381 } 2382 2383 const Stmt *PathSensitiveBugReport::getStmt() const { 2384 if (!ErrorNode) 2385 return nullptr; 2386 2387 ProgramPoint ProgP = ErrorNode->getLocation(); 2388 const Stmt *S = nullptr; 2389 2390 if (std::optional<BlockEntrance> BE = ProgP.getAs<BlockEntrance>()) { 2391 CFGBlock &Exit = ProgP.getLocationContext()->getCFG()->getExit(); 2392 if (BE->getBlock() == &Exit) 2393 S = ErrorNode->getPreviousStmtForDiagnostics(); 2394 } 2395 if (!S) 2396 S = ErrorNode->getStmtForDiagnostics(); 2397 2398 return S; 2399 } 2400 2401 ArrayRef<SourceRange> 2402 PathSensitiveBugReport::getRanges() const { 2403 // If no custom ranges, add the range of the statement corresponding to 2404 // the error node. 2405 if (Ranges.empty() && isa_and_nonnull<Expr>(getStmt())) 2406 return ErrorNodeRange; 2407 2408 return Ranges; 2409 } 2410 2411 PathDiagnosticLocation 2412 PathSensitiveBugReport::getLocation() const { 2413 assert(ErrorNode && "Cannot create a location with a null node."); 2414 const Stmt *S = ErrorNode->getStmtForDiagnostics(); 2415 ProgramPoint P = ErrorNode->getLocation(); 2416 const LocationContext *LC = P.getLocationContext(); 2417 SourceManager &SM = 2418 ErrorNode->getState()->getStateManager().getContext().getSourceManager(); 2419 2420 if (!S) { 2421 // If this is an implicit call, return the implicit call point location. 2422 if (std::optional<PreImplicitCall> PIE = P.getAs<PreImplicitCall>()) 2423 return PathDiagnosticLocation(PIE->getLocation(), SM); 2424 if (auto FE = P.getAs<FunctionExitPoint>()) { 2425 if (const ReturnStmt *RS = FE->getStmt()) 2426 return PathDiagnosticLocation::createBegin(RS, SM, LC); 2427 } 2428 S = ErrorNode->getNextStmtForDiagnostics(); 2429 } 2430 2431 if (S) { 2432 // For member expressions, return the location of the '.' or '->'. 2433 if (const auto *ME = dyn_cast<MemberExpr>(S)) 2434 return PathDiagnosticLocation::createMemberLoc(ME, SM); 2435 2436 // For binary operators, return the location of the operator. 2437 if (const auto *B = dyn_cast<BinaryOperator>(S)) 2438 return PathDiagnosticLocation::createOperatorLoc(B, SM); 2439 2440 if (P.getAs<PostStmtPurgeDeadSymbols>()) 2441 return PathDiagnosticLocation::createEnd(S, SM, LC); 2442 2443 if (S->getBeginLoc().isValid()) 2444 return PathDiagnosticLocation(S, SM, LC); 2445 2446 return PathDiagnosticLocation( 2447 PathDiagnosticLocation::getValidSourceLocation(S, LC), SM); 2448 } 2449 2450 return PathDiagnosticLocation::createDeclEnd(ErrorNode->getLocationContext(), 2451 SM); 2452 } 2453 2454 //===----------------------------------------------------------------------===// 2455 // Methods for BugReporter and subclasses. 2456 //===----------------------------------------------------------------------===// 2457 2458 const ExplodedGraph &PathSensitiveBugReporter::getGraph() const { 2459 return Eng.getGraph(); 2460 } 2461 2462 ProgramStateManager &PathSensitiveBugReporter::getStateManager() const { 2463 return Eng.getStateManager(); 2464 } 2465 2466 BugReporter::BugReporter(BugReporterData &d) : D(d) {} 2467 BugReporter::~BugReporter() { 2468 // Make sure reports are flushed. 2469 assert(StrBugTypes.empty() && 2470 "Destroying BugReporter before diagnostics are emitted!"); 2471 2472 // Free the bug reports we are tracking. 2473 for (const auto I : EQClassesVector) 2474 delete I; 2475 } 2476 2477 void BugReporter::FlushReports() { 2478 // We need to flush reports in deterministic order to ensure the order 2479 // of the reports is consistent between runs. 2480 for (const auto EQ : EQClassesVector) 2481 FlushReport(*EQ); 2482 2483 // BugReporter owns and deletes only BugTypes created implicitly through 2484 // EmitBasicReport. 2485 // FIXME: There are leaks from checkers that assume that the BugTypes they 2486 // create will be destroyed by the BugReporter. 2487 StrBugTypes.clear(); 2488 } 2489 2490 //===----------------------------------------------------------------------===// 2491 // PathDiagnostics generation. 2492 //===----------------------------------------------------------------------===// 2493 2494 namespace { 2495 2496 /// A wrapper around an ExplodedGraph that contains a single path from the root 2497 /// to the error node. 2498 class BugPathInfo { 2499 public: 2500 std::unique_ptr<ExplodedGraph> BugPath; 2501 PathSensitiveBugReport *Report; 2502 const ExplodedNode *ErrorNode; 2503 }; 2504 2505 /// A wrapper around an ExplodedGraph whose leafs are all error nodes. Can 2506 /// conveniently retrieve bug paths from a single error node to the root. 2507 class BugPathGetter { 2508 std::unique_ptr<ExplodedGraph> TrimmedGraph; 2509 2510 using PriorityMapTy = llvm::DenseMap<const ExplodedNode *, unsigned>; 2511 2512 /// Assign each node with its distance from the root. 2513 PriorityMapTy PriorityMap; 2514 2515 /// Since the getErrorNode() or BugReport refers to the original ExplodedGraph, 2516 /// we need to pair it to the error node of the constructed trimmed graph. 2517 using ReportNewNodePair = 2518 std::pair<PathSensitiveBugReport *, const ExplodedNode *>; 2519 SmallVector<ReportNewNodePair, 32> ReportNodes; 2520 2521 BugPathInfo CurrentBugPath; 2522 2523 /// A helper class for sorting ExplodedNodes by priority. 2524 template <bool Descending> 2525 class PriorityCompare { 2526 const PriorityMapTy &PriorityMap; 2527 2528 public: 2529 PriorityCompare(const PriorityMapTy &M) : PriorityMap(M) {} 2530 2531 bool operator()(const ExplodedNode *LHS, const ExplodedNode *RHS) const { 2532 PriorityMapTy::const_iterator LI = PriorityMap.find(LHS); 2533 PriorityMapTy::const_iterator RI = PriorityMap.find(RHS); 2534 PriorityMapTy::const_iterator E = PriorityMap.end(); 2535 2536 if (LI == E) 2537 return Descending; 2538 if (RI == E) 2539 return !Descending; 2540 2541 return Descending ? LI->second > RI->second 2542 : LI->second < RI->second; 2543 } 2544 2545 bool operator()(const ReportNewNodePair &LHS, 2546 const ReportNewNodePair &RHS) const { 2547 return (*this)(LHS.second, RHS.second); 2548 } 2549 }; 2550 2551 public: 2552 BugPathGetter(const ExplodedGraph *OriginalGraph, 2553 ArrayRef<PathSensitiveBugReport *> &bugReports); 2554 2555 BugPathInfo *getNextBugPath(); 2556 }; 2557 2558 } // namespace 2559 2560 BugPathGetter::BugPathGetter(const ExplodedGraph *OriginalGraph, 2561 ArrayRef<PathSensitiveBugReport *> &bugReports) { 2562 SmallVector<const ExplodedNode *, 32> Nodes; 2563 for (const auto I : bugReports) { 2564 assert(I->isValid() && 2565 "We only allow BugReporterVisitors and BugReporter itself to " 2566 "invalidate reports!"); 2567 Nodes.emplace_back(I->getErrorNode()); 2568 } 2569 2570 // The trimmed graph is created in the body of the constructor to ensure 2571 // that the DenseMaps have been initialized already. 2572 InterExplodedGraphMap ForwardMap; 2573 TrimmedGraph = OriginalGraph->trim(Nodes, &ForwardMap); 2574 2575 // Find the (first) error node in the trimmed graph. We just need to consult 2576 // the node map which maps from nodes in the original graph to nodes 2577 // in the new graph. 2578 llvm::SmallPtrSet<const ExplodedNode *, 32> RemainingNodes; 2579 2580 for (PathSensitiveBugReport *Report : bugReports) { 2581 const ExplodedNode *NewNode = ForwardMap.lookup(Report->getErrorNode()); 2582 assert(NewNode && 2583 "Failed to construct a trimmed graph that contains this error " 2584 "node!"); 2585 ReportNodes.emplace_back(Report, NewNode); 2586 RemainingNodes.insert(NewNode); 2587 } 2588 2589 assert(!RemainingNodes.empty() && "No error node found in the trimmed graph"); 2590 2591 // Perform a forward BFS to find all the shortest paths. 2592 std::queue<const ExplodedNode *> WS; 2593 2594 assert(TrimmedGraph->num_roots() == 1); 2595 WS.push(*TrimmedGraph->roots_begin()); 2596 unsigned Priority = 0; 2597 2598 while (!WS.empty()) { 2599 const ExplodedNode *Node = WS.front(); 2600 WS.pop(); 2601 2602 PriorityMapTy::iterator PriorityEntry; 2603 bool IsNew; 2604 std::tie(PriorityEntry, IsNew) = PriorityMap.insert({Node, Priority}); 2605 ++Priority; 2606 2607 if (!IsNew) { 2608 assert(PriorityEntry->second <= Priority); 2609 continue; 2610 } 2611 2612 if (RemainingNodes.erase(Node)) 2613 if (RemainingNodes.empty()) 2614 break; 2615 2616 for (const ExplodedNode *Succ : Node->succs()) 2617 WS.push(Succ); 2618 } 2619 2620 // Sort the error paths from longest to shortest. 2621 llvm::sort(ReportNodes, PriorityCompare<true>(PriorityMap)); 2622 } 2623 2624 BugPathInfo *BugPathGetter::getNextBugPath() { 2625 if (ReportNodes.empty()) 2626 return nullptr; 2627 2628 const ExplodedNode *OrigN; 2629 std::tie(CurrentBugPath.Report, OrigN) = ReportNodes.pop_back_val(); 2630 assert(PriorityMap.find(OrigN) != PriorityMap.end() && 2631 "error node not accessible from root"); 2632 2633 // Create a new graph with a single path. This is the graph that will be 2634 // returned to the caller. 2635 auto GNew = std::make_unique<ExplodedGraph>(); 2636 2637 // Now walk from the error node up the BFS path, always taking the 2638 // predeccessor with the lowest number. 2639 ExplodedNode *Succ = nullptr; 2640 while (true) { 2641 // Create the equivalent node in the new graph with the same state 2642 // and location. 2643 ExplodedNode *NewN = GNew->createUncachedNode( 2644 OrigN->getLocation(), OrigN->getState(), 2645 OrigN->getID(), OrigN->isSink()); 2646 2647 // Link up the new node with the previous node. 2648 if (Succ) 2649 Succ->addPredecessor(NewN, *GNew); 2650 else 2651 CurrentBugPath.ErrorNode = NewN; 2652 2653 Succ = NewN; 2654 2655 // Are we at the final node? 2656 if (OrigN->pred_empty()) { 2657 GNew->addRoot(NewN); 2658 break; 2659 } 2660 2661 // Find the next predeccessor node. We choose the node that is marked 2662 // with the lowest BFS number. 2663 OrigN = *std::min_element(OrigN->pred_begin(), OrigN->pred_end(), 2664 PriorityCompare<false>(PriorityMap)); 2665 } 2666 2667 CurrentBugPath.BugPath = std::move(GNew); 2668 2669 return &CurrentBugPath; 2670 } 2671 2672 /// CompactMacroExpandedPieces - This function postprocesses a PathDiagnostic 2673 /// object and collapses PathDiagosticPieces that are expanded by macros. 2674 static void CompactMacroExpandedPieces(PathPieces &path, 2675 const SourceManager& SM) { 2676 using MacroStackTy = std::vector< 2677 std::pair<std::shared_ptr<PathDiagnosticMacroPiece>, SourceLocation>>; 2678 2679 using PiecesTy = std::vector<PathDiagnosticPieceRef>; 2680 2681 MacroStackTy MacroStack; 2682 PiecesTy Pieces; 2683 2684 for (PathPieces::const_iterator I = path.begin(), E = path.end(); 2685 I != E; ++I) { 2686 const auto &piece = *I; 2687 2688 // Recursively compact calls. 2689 if (auto *call = dyn_cast<PathDiagnosticCallPiece>(&*piece)) { 2690 CompactMacroExpandedPieces(call->path, SM); 2691 } 2692 2693 // Get the location of the PathDiagnosticPiece. 2694 const FullSourceLoc Loc = piece->getLocation().asLocation(); 2695 2696 // Determine the instantiation location, which is the location we group 2697 // related PathDiagnosticPieces. 2698 SourceLocation InstantiationLoc = Loc.isMacroID() ? 2699 SM.getExpansionLoc(Loc) : 2700 SourceLocation(); 2701 2702 if (Loc.isFileID()) { 2703 MacroStack.clear(); 2704 Pieces.push_back(piece); 2705 continue; 2706 } 2707 2708 assert(Loc.isMacroID()); 2709 2710 // Is the PathDiagnosticPiece within the same macro group? 2711 if (!MacroStack.empty() && InstantiationLoc == MacroStack.back().second) { 2712 MacroStack.back().first->subPieces.push_back(piece); 2713 continue; 2714 } 2715 2716 // We aren't in the same group. Are we descending into a new macro 2717 // or are part of an old one? 2718 std::shared_ptr<PathDiagnosticMacroPiece> MacroGroup; 2719 2720 SourceLocation ParentInstantiationLoc = InstantiationLoc.isMacroID() ? 2721 SM.getExpansionLoc(Loc) : 2722 SourceLocation(); 2723 2724 // Walk the entire macro stack. 2725 while (!MacroStack.empty()) { 2726 if (InstantiationLoc == MacroStack.back().second) { 2727 MacroGroup = MacroStack.back().first; 2728 break; 2729 } 2730 2731 if (ParentInstantiationLoc == MacroStack.back().second) { 2732 MacroGroup = MacroStack.back().first; 2733 break; 2734 } 2735 2736 MacroStack.pop_back(); 2737 } 2738 2739 if (!MacroGroup || ParentInstantiationLoc == MacroStack.back().second) { 2740 // Create a new macro group and add it to the stack. 2741 auto NewGroup = std::make_shared<PathDiagnosticMacroPiece>( 2742 PathDiagnosticLocation::createSingleLocation(piece->getLocation())); 2743 2744 if (MacroGroup) 2745 MacroGroup->subPieces.push_back(NewGroup); 2746 else { 2747 assert(InstantiationLoc.isFileID()); 2748 Pieces.push_back(NewGroup); 2749 } 2750 2751 MacroGroup = NewGroup; 2752 MacroStack.push_back(std::make_pair(MacroGroup, InstantiationLoc)); 2753 } 2754 2755 // Finally, add the PathDiagnosticPiece to the group. 2756 MacroGroup->subPieces.push_back(piece); 2757 } 2758 2759 // Now take the pieces and construct a new PathDiagnostic. 2760 path.clear(); 2761 2762 path.insert(path.end(), Pieces.begin(), Pieces.end()); 2763 } 2764 2765 /// Generate notes from all visitors. 2766 /// Notes associated with @c ErrorNode are generated using 2767 /// @c getEndPath, and the rest are generated with @c VisitNode. 2768 static std::unique_ptr<VisitorsDiagnosticsTy> 2769 generateVisitorsDiagnostics(PathSensitiveBugReport *R, 2770 const ExplodedNode *ErrorNode, 2771 BugReporterContext &BRC) { 2772 std::unique_ptr<VisitorsDiagnosticsTy> Notes = 2773 std::make_unique<VisitorsDiagnosticsTy>(); 2774 PathSensitiveBugReport::VisitorList visitors; 2775 2776 // Run visitors on all nodes starting from the node *before* the last one. 2777 // The last node is reserved for notes generated with @c getEndPath. 2778 const ExplodedNode *NextNode = ErrorNode->getFirstPred(); 2779 while (NextNode) { 2780 2781 // At each iteration, move all visitors from report to visitor list. This is 2782 // important, because the Profile() functions of the visitors make sure that 2783 // a visitor isn't added multiple times for the same node, but it's fine 2784 // to add the a visitor with Profile() for different nodes (e.g. tracking 2785 // a region at different points of the symbolic execution). 2786 for (std::unique_ptr<BugReporterVisitor> &Visitor : R->visitors()) 2787 visitors.push_back(std::move(Visitor)); 2788 2789 R->clearVisitors(); 2790 2791 const ExplodedNode *Pred = NextNode->getFirstPred(); 2792 if (!Pred) { 2793 PathDiagnosticPieceRef LastPiece; 2794 for (auto &V : visitors) { 2795 V->finalizeVisitor(BRC, ErrorNode, *R); 2796 2797 if (auto Piece = V->getEndPath(BRC, ErrorNode, *R)) { 2798 assert(!LastPiece && 2799 "There can only be one final piece in a diagnostic."); 2800 assert(Piece->getKind() == PathDiagnosticPiece::Kind::Event && 2801 "The final piece must contain a message!"); 2802 LastPiece = std::move(Piece); 2803 (*Notes)[ErrorNode].push_back(LastPiece); 2804 } 2805 } 2806 break; 2807 } 2808 2809 for (auto &V : visitors) { 2810 auto P = V->VisitNode(NextNode, BRC, *R); 2811 if (P) 2812 (*Notes)[NextNode].push_back(std::move(P)); 2813 } 2814 2815 if (!R->isValid()) 2816 break; 2817 2818 NextNode = Pred; 2819 } 2820 2821 return Notes; 2822 } 2823 2824 std::optional<PathDiagnosticBuilder> PathDiagnosticBuilder::findValidReport( 2825 ArrayRef<PathSensitiveBugReport *> &bugReports, 2826 PathSensitiveBugReporter &Reporter) { 2827 2828 BugPathGetter BugGraph(&Reporter.getGraph(), bugReports); 2829 2830 while (BugPathInfo *BugPath = BugGraph.getNextBugPath()) { 2831 // Find the BugReport with the original location. 2832 PathSensitiveBugReport *R = BugPath->Report; 2833 assert(R && "No original report found for sliced graph."); 2834 assert(R->isValid() && "Report selected by trimmed graph marked invalid."); 2835 const ExplodedNode *ErrorNode = BugPath->ErrorNode; 2836 2837 // Register refutation visitors first, if they mark the bug invalid no 2838 // further analysis is required 2839 R->addVisitor<LikelyFalsePositiveSuppressionBRVisitor>(); 2840 2841 // Register additional node visitors. 2842 R->addVisitor<NilReceiverBRVisitor>(); 2843 R->addVisitor<ConditionBRVisitor>(); 2844 R->addVisitor<TagVisitor>(); 2845 2846 BugReporterContext BRC(Reporter); 2847 2848 // Run all visitors on a given graph, once. 2849 std::unique_ptr<VisitorsDiagnosticsTy> visitorNotes = 2850 generateVisitorsDiagnostics(R, ErrorNode, BRC); 2851 2852 if (R->isValid()) { 2853 if (Reporter.getAnalyzerOptions().ShouldCrosscheckWithZ3) { 2854 // If crosscheck is enabled, remove all visitors, add the refutation 2855 // visitor and check again 2856 R->clearVisitors(); 2857 R->addVisitor<FalsePositiveRefutationBRVisitor>(); 2858 2859 // We don't overwrite the notes inserted by other visitors because the 2860 // refutation manager does not add any new note to the path 2861 generateVisitorsDiagnostics(R, BugPath->ErrorNode, BRC); 2862 } 2863 2864 // Check if the bug is still valid 2865 if (R->isValid()) 2866 return PathDiagnosticBuilder( 2867 std::move(BRC), std::move(BugPath->BugPath), BugPath->Report, 2868 BugPath->ErrorNode, std::move(visitorNotes)); 2869 } 2870 } 2871 2872 return {}; 2873 } 2874 2875 std::unique_ptr<DiagnosticForConsumerMapTy> 2876 PathSensitiveBugReporter::generatePathDiagnostics( 2877 ArrayRef<PathDiagnosticConsumer *> consumers, 2878 ArrayRef<PathSensitiveBugReport *> &bugReports) { 2879 assert(!bugReports.empty()); 2880 2881 auto Out = std::make_unique<DiagnosticForConsumerMapTy>(); 2882 2883 std::optional<PathDiagnosticBuilder> PDB = 2884 PathDiagnosticBuilder::findValidReport(bugReports, *this); 2885 2886 if (PDB) { 2887 for (PathDiagnosticConsumer *PC : consumers) { 2888 if (std::unique_ptr<PathDiagnostic> PD = PDB->generate(PC)) { 2889 (*Out)[PC] = std::move(PD); 2890 } 2891 } 2892 } 2893 2894 return Out; 2895 } 2896 2897 void BugReporter::emitReport(std::unique_ptr<BugReport> R) { 2898 bool ValidSourceLoc = R->getLocation().isValid(); 2899 assert(ValidSourceLoc); 2900 // If we mess up in a release build, we'd still prefer to just drop the bug 2901 // instead of trying to go on. 2902 if (!ValidSourceLoc) 2903 return; 2904 2905 // Compute the bug report's hash to determine its equivalence class. 2906 llvm::FoldingSetNodeID ID; 2907 R->Profile(ID); 2908 2909 // Lookup the equivance class. If there isn't one, create it. 2910 void *InsertPos; 2911 BugReportEquivClass* EQ = EQClasses.FindNodeOrInsertPos(ID, InsertPos); 2912 2913 if (!EQ) { 2914 EQ = new BugReportEquivClass(std::move(R)); 2915 EQClasses.InsertNode(EQ, InsertPos); 2916 EQClassesVector.push_back(EQ); 2917 } else 2918 EQ->AddReport(std::move(R)); 2919 } 2920 2921 void PathSensitiveBugReporter::emitReport(std::unique_ptr<BugReport> R) { 2922 if (auto PR = dyn_cast<PathSensitiveBugReport>(R.get())) 2923 if (const ExplodedNode *E = PR->getErrorNode()) { 2924 // An error node must either be a sink or have a tag, otherwise 2925 // it could get reclaimed before the path diagnostic is created. 2926 assert((E->isSink() || E->getLocation().getTag()) && 2927 "Error node must either be a sink or have a tag"); 2928 2929 const AnalysisDeclContext *DeclCtx = 2930 E->getLocationContext()->getAnalysisDeclContext(); 2931 // The source of autosynthesized body can be handcrafted AST or a model 2932 // file. The locations from handcrafted ASTs have no valid source 2933 // locations and have to be discarded. Locations from model files should 2934 // be preserved for processing and reporting. 2935 if (DeclCtx->isBodyAutosynthesized() && 2936 !DeclCtx->isBodyAutosynthesizedFromModelFile()) 2937 return; 2938 } 2939 2940 BugReporter::emitReport(std::move(R)); 2941 } 2942 2943 //===----------------------------------------------------------------------===// 2944 // Emitting reports in equivalence classes. 2945 //===----------------------------------------------------------------------===// 2946 2947 namespace { 2948 2949 struct FRIEC_WLItem { 2950 const ExplodedNode *N; 2951 ExplodedNode::const_succ_iterator I, E; 2952 2953 FRIEC_WLItem(const ExplodedNode *n) 2954 : N(n), I(N->succ_begin()), E(N->succ_end()) {} 2955 }; 2956 2957 } // namespace 2958 2959 BugReport *PathSensitiveBugReporter::findReportInEquivalenceClass( 2960 BugReportEquivClass &EQ, SmallVectorImpl<BugReport *> &bugReports) { 2961 // If we don't need to suppress any of the nodes because they are 2962 // post-dominated by a sink, simply add all the nodes in the equivalence class 2963 // to 'Nodes'. Any of the reports will serve as a "representative" report. 2964 assert(EQ.getReports().size() > 0); 2965 const BugType& BT = EQ.getReports()[0]->getBugType(); 2966 if (!BT.isSuppressOnSink()) { 2967 BugReport *R = EQ.getReports()[0].get(); 2968 for (auto &J : EQ.getReports()) { 2969 if (auto *PR = dyn_cast<PathSensitiveBugReport>(J.get())) { 2970 R = PR; 2971 bugReports.push_back(PR); 2972 } 2973 } 2974 return R; 2975 } 2976 2977 // For bug reports that should be suppressed when all paths are post-dominated 2978 // by a sink node, iterate through the reports in the equivalence class 2979 // until we find one that isn't post-dominated (if one exists). We use a 2980 // DFS traversal of the ExplodedGraph to find a non-sink node. We could write 2981 // this as a recursive function, but we don't want to risk blowing out the 2982 // stack for very long paths. 2983 BugReport *exampleReport = nullptr; 2984 2985 for (const auto &I: EQ.getReports()) { 2986 auto *R = dyn_cast<PathSensitiveBugReport>(I.get()); 2987 if (!R) 2988 continue; 2989 2990 const ExplodedNode *errorNode = R->getErrorNode(); 2991 if (errorNode->isSink()) { 2992 llvm_unreachable( 2993 "BugType::isSuppressSink() should not be 'true' for sink end nodes"); 2994 } 2995 // No successors? By definition this nodes isn't post-dominated by a sink. 2996 if (errorNode->succ_empty()) { 2997 bugReports.push_back(R); 2998 if (!exampleReport) 2999 exampleReport = R; 3000 continue; 3001 } 3002 3003 // See if we are in a no-return CFG block. If so, treat this similarly 3004 // to being post-dominated by a sink. This works better when the analysis 3005 // is incomplete and we have never reached the no-return function call(s) 3006 // that we'd inevitably bump into on this path. 3007 if (const CFGBlock *ErrorB = errorNode->getCFGBlock()) 3008 if (ErrorB->isInevitablySinking()) 3009 continue; 3010 3011 // At this point we know that 'N' is not a sink and it has at least one 3012 // successor. Use a DFS worklist to find a non-sink end-of-path node. 3013 using WLItem = FRIEC_WLItem; 3014 using DFSWorkList = SmallVector<WLItem, 10>; 3015 3016 llvm::DenseMap<const ExplodedNode *, unsigned> Visited; 3017 3018 DFSWorkList WL; 3019 WL.push_back(errorNode); 3020 Visited[errorNode] = 1; 3021 3022 while (!WL.empty()) { 3023 WLItem &WI = WL.back(); 3024 assert(!WI.N->succ_empty()); 3025 3026 for (; WI.I != WI.E; ++WI.I) { 3027 const ExplodedNode *Succ = *WI.I; 3028 // End-of-path node? 3029 if (Succ->succ_empty()) { 3030 // If we found an end-of-path node that is not a sink. 3031 if (!Succ->isSink()) { 3032 bugReports.push_back(R); 3033 if (!exampleReport) 3034 exampleReport = R; 3035 WL.clear(); 3036 break; 3037 } 3038 // Found a sink? Continue on to the next successor. 3039 continue; 3040 } 3041 // Mark the successor as visited. If it hasn't been explored, 3042 // enqueue it to the DFS worklist. 3043 unsigned &mark = Visited[Succ]; 3044 if (!mark) { 3045 mark = 1; 3046 WL.push_back(Succ); 3047 break; 3048 } 3049 } 3050 3051 // The worklist may have been cleared at this point. First 3052 // check if it is empty before checking the last item. 3053 if (!WL.empty() && &WL.back() == &WI) 3054 WL.pop_back(); 3055 } 3056 } 3057 3058 // ExampleReport will be NULL if all the nodes in the equivalence class 3059 // were post-dominated by sinks. 3060 return exampleReport; 3061 } 3062 3063 void BugReporter::FlushReport(BugReportEquivClass& EQ) { 3064 SmallVector<BugReport*, 10> bugReports; 3065 BugReport *report = findReportInEquivalenceClass(EQ, bugReports); 3066 if (!report) 3067 return; 3068 3069 // See whether we need to silence the checker/package. 3070 for (const std::string &CheckerOrPackage : 3071 getAnalyzerOptions().SilencedCheckersAndPackages) { 3072 if (report->getBugType().getCheckerName().startswith( 3073 CheckerOrPackage)) 3074 return; 3075 } 3076 3077 ArrayRef<PathDiagnosticConsumer*> Consumers = getPathDiagnosticConsumers(); 3078 std::unique_ptr<DiagnosticForConsumerMapTy> Diagnostics = 3079 generateDiagnosticForConsumerMap(report, Consumers, bugReports); 3080 3081 for (auto &P : *Diagnostics) { 3082 PathDiagnosticConsumer *Consumer = P.first; 3083 std::unique_ptr<PathDiagnostic> &PD = P.second; 3084 3085 // If the path is empty, generate a single step path with the location 3086 // of the issue. 3087 if (PD->path.empty()) { 3088 PathDiagnosticLocation L = report->getLocation(); 3089 auto piece = std::make_unique<PathDiagnosticEventPiece>( 3090 L, report->getDescription()); 3091 for (SourceRange Range : report->getRanges()) 3092 piece->addRange(Range); 3093 PD->setEndOfPath(std::move(piece)); 3094 } 3095 3096 PathPieces &Pieces = PD->getMutablePieces(); 3097 if (getAnalyzerOptions().ShouldDisplayNotesAsEvents) { 3098 // For path diagnostic consumers that don't support extra notes, 3099 // we may optionally convert those to path notes. 3100 for (const auto &I : llvm::reverse(report->getNotes())) { 3101 PathDiagnosticNotePiece *Piece = I.get(); 3102 auto ConvertedPiece = std::make_shared<PathDiagnosticEventPiece>( 3103 Piece->getLocation(), Piece->getString()); 3104 for (const auto &R: Piece->getRanges()) 3105 ConvertedPiece->addRange(R); 3106 3107 Pieces.push_front(std::move(ConvertedPiece)); 3108 } 3109 } else { 3110 for (const auto &I : llvm::reverse(report->getNotes())) 3111 Pieces.push_front(I); 3112 } 3113 3114 for (const auto &I : report->getFixits()) 3115 Pieces.back()->addFixit(I); 3116 3117 updateExecutedLinesWithDiagnosticPieces(*PD); 3118 Consumer->HandlePathDiagnostic(std::move(PD)); 3119 } 3120 } 3121 3122 /// Insert all lines participating in the function signature \p Signature 3123 /// into \p ExecutedLines. 3124 static void populateExecutedLinesWithFunctionSignature( 3125 const Decl *Signature, const SourceManager &SM, 3126 FilesToLineNumsMap &ExecutedLines) { 3127 SourceRange SignatureSourceRange; 3128 const Stmt* Body = Signature->getBody(); 3129 if (const auto FD = dyn_cast<FunctionDecl>(Signature)) { 3130 SignatureSourceRange = FD->getSourceRange(); 3131 } else if (const auto OD = dyn_cast<ObjCMethodDecl>(Signature)) { 3132 SignatureSourceRange = OD->getSourceRange(); 3133 } else { 3134 return; 3135 } 3136 SourceLocation Start = SignatureSourceRange.getBegin(); 3137 SourceLocation End = Body ? Body->getSourceRange().getBegin() 3138 : SignatureSourceRange.getEnd(); 3139 if (!Start.isValid() || !End.isValid()) 3140 return; 3141 unsigned StartLine = SM.getExpansionLineNumber(Start); 3142 unsigned EndLine = SM.getExpansionLineNumber(End); 3143 3144 FileID FID = SM.getFileID(SM.getExpansionLoc(Start)); 3145 for (unsigned Line = StartLine; Line <= EndLine; Line++) 3146 ExecutedLines[FID].insert(Line); 3147 } 3148 3149 static void populateExecutedLinesWithStmt( 3150 const Stmt *S, const SourceManager &SM, 3151 FilesToLineNumsMap &ExecutedLines) { 3152 SourceLocation Loc = S->getSourceRange().getBegin(); 3153 if (!Loc.isValid()) 3154 return; 3155 SourceLocation ExpansionLoc = SM.getExpansionLoc(Loc); 3156 FileID FID = SM.getFileID(ExpansionLoc); 3157 unsigned LineNo = SM.getExpansionLineNumber(ExpansionLoc); 3158 ExecutedLines[FID].insert(LineNo); 3159 } 3160 3161 /// \return all executed lines including function signatures on the path 3162 /// starting from \p N. 3163 static std::unique_ptr<FilesToLineNumsMap> 3164 findExecutedLines(const SourceManager &SM, const ExplodedNode *N) { 3165 auto ExecutedLines = std::make_unique<FilesToLineNumsMap>(); 3166 3167 while (N) { 3168 if (N->getFirstPred() == nullptr) { 3169 // First node: show signature of the entrance point. 3170 const Decl *D = N->getLocationContext()->getDecl(); 3171 populateExecutedLinesWithFunctionSignature(D, SM, *ExecutedLines); 3172 } else if (auto CE = N->getLocationAs<CallEnter>()) { 3173 // Inlined function: show signature. 3174 const Decl* D = CE->getCalleeContext()->getDecl(); 3175 populateExecutedLinesWithFunctionSignature(D, SM, *ExecutedLines); 3176 } else if (const Stmt *S = N->getStmtForDiagnostics()) { 3177 populateExecutedLinesWithStmt(S, SM, *ExecutedLines); 3178 3179 // Show extra context for some parent kinds. 3180 const Stmt *P = N->getParentMap().getParent(S); 3181 3182 // The path exploration can die before the node with the associated 3183 // return statement is generated, but we do want to show the whole 3184 // return. 3185 if (const auto *RS = dyn_cast_or_null<ReturnStmt>(P)) { 3186 populateExecutedLinesWithStmt(RS, SM, *ExecutedLines); 3187 P = N->getParentMap().getParent(RS); 3188 } 3189 3190 if (isa_and_nonnull<SwitchCase, LabelStmt>(P)) 3191 populateExecutedLinesWithStmt(P, SM, *ExecutedLines); 3192 } 3193 3194 N = N->getFirstPred(); 3195 } 3196 return ExecutedLines; 3197 } 3198 3199 std::unique_ptr<DiagnosticForConsumerMapTy> 3200 BugReporter::generateDiagnosticForConsumerMap( 3201 BugReport *exampleReport, ArrayRef<PathDiagnosticConsumer *> consumers, 3202 ArrayRef<BugReport *> bugReports) { 3203 auto *basicReport = cast<BasicBugReport>(exampleReport); 3204 auto Out = std::make_unique<DiagnosticForConsumerMapTy>(); 3205 for (auto *Consumer : consumers) 3206 (*Out)[Consumer] = generateDiagnosticForBasicReport(basicReport); 3207 return Out; 3208 } 3209 3210 static PathDiagnosticCallPiece * 3211 getFirstStackedCallToHeaderFile(PathDiagnosticCallPiece *CP, 3212 const SourceManager &SMgr) { 3213 SourceLocation CallLoc = CP->callEnter.asLocation(); 3214 3215 // If the call is within a macro, don't do anything (for now). 3216 if (CallLoc.isMacroID()) 3217 return nullptr; 3218 3219 assert(AnalysisManager::isInCodeFile(CallLoc, SMgr) && 3220 "The call piece should not be in a header file."); 3221 3222 // Check if CP represents a path through a function outside of the main file. 3223 if (!AnalysisManager::isInCodeFile(CP->callEnterWithin.asLocation(), SMgr)) 3224 return CP; 3225 3226 const PathPieces &Path = CP->path; 3227 if (Path.empty()) 3228 return nullptr; 3229 3230 // Check if the last piece in the callee path is a call to a function outside 3231 // of the main file. 3232 if (auto *CPInner = dyn_cast<PathDiagnosticCallPiece>(Path.back().get())) 3233 return getFirstStackedCallToHeaderFile(CPInner, SMgr); 3234 3235 // Otherwise, the last piece is in the main file. 3236 return nullptr; 3237 } 3238 3239 static void resetDiagnosticLocationToMainFile(PathDiagnostic &PD) { 3240 if (PD.path.empty()) 3241 return; 3242 3243 PathDiagnosticPiece *LastP = PD.path.back().get(); 3244 assert(LastP); 3245 const SourceManager &SMgr = LastP->getLocation().getManager(); 3246 3247 // We only need to check if the report ends inside headers, if the last piece 3248 // is a call piece. 3249 if (auto *CP = dyn_cast<PathDiagnosticCallPiece>(LastP)) { 3250 CP = getFirstStackedCallToHeaderFile(CP, SMgr); 3251 if (CP) { 3252 // Mark the piece. 3253 CP->setAsLastInMainSourceFile(); 3254 3255 // Update the path diagnostic message. 3256 const auto *ND = dyn_cast<NamedDecl>(CP->getCallee()); 3257 if (ND) { 3258 SmallString<200> buf; 3259 llvm::raw_svector_ostream os(buf); 3260 os << " (within a call to '" << ND->getDeclName() << "')"; 3261 PD.appendToDesc(os.str()); 3262 } 3263 3264 // Reset the report containing declaration and location. 3265 PD.setDeclWithIssue(CP->getCaller()); 3266 PD.setLocation(CP->getLocation()); 3267 3268 return; 3269 } 3270 } 3271 } 3272 3273 3274 3275 std::unique_ptr<DiagnosticForConsumerMapTy> 3276 PathSensitiveBugReporter::generateDiagnosticForConsumerMap( 3277 BugReport *exampleReport, ArrayRef<PathDiagnosticConsumer *> consumers, 3278 ArrayRef<BugReport *> bugReports) { 3279 std::vector<BasicBugReport *> BasicBugReports; 3280 std::vector<PathSensitiveBugReport *> PathSensitiveBugReports; 3281 if (isa<BasicBugReport>(exampleReport)) 3282 return BugReporter::generateDiagnosticForConsumerMap(exampleReport, 3283 consumers, bugReports); 3284 3285 // Generate the full path sensitive diagnostic, using the generation scheme 3286 // specified by the PathDiagnosticConsumer. Note that we have to generate 3287 // path diagnostics even for consumers which do not support paths, because 3288 // the BugReporterVisitors may mark this bug as a false positive. 3289 assert(!bugReports.empty()); 3290 MaxBugClassSize.updateMax(bugReports.size()); 3291 3292 // Avoid copying the whole array because there may be a lot of reports. 3293 ArrayRef<PathSensitiveBugReport *> convertedArrayOfReports( 3294 reinterpret_cast<PathSensitiveBugReport *const *>(&*bugReports.begin()), 3295 reinterpret_cast<PathSensitiveBugReport *const *>(&*bugReports.end())); 3296 std::unique_ptr<DiagnosticForConsumerMapTy> Out = generatePathDiagnostics( 3297 consumers, convertedArrayOfReports); 3298 3299 if (Out->empty()) 3300 return Out; 3301 3302 MaxValidBugClassSize.updateMax(bugReports.size()); 3303 3304 // Examine the report and see if the last piece is in a header. Reset the 3305 // report location to the last piece in the main source file. 3306 const AnalyzerOptions &Opts = getAnalyzerOptions(); 3307 for (auto const &P : *Out) 3308 if (Opts.ShouldReportIssuesInMainSourceFile && !Opts.AnalyzeAll) 3309 resetDiagnosticLocationToMainFile(*P.second); 3310 3311 return Out; 3312 } 3313 3314 void BugReporter::EmitBasicReport(const Decl *DeclWithIssue, 3315 const CheckerBase *Checker, StringRef Name, 3316 StringRef Category, StringRef Str, 3317 PathDiagnosticLocation Loc, 3318 ArrayRef<SourceRange> Ranges, 3319 ArrayRef<FixItHint> Fixits) { 3320 EmitBasicReport(DeclWithIssue, Checker->getCheckerName(), Name, Category, Str, 3321 Loc, Ranges, Fixits); 3322 } 3323 3324 void BugReporter::EmitBasicReport(const Decl *DeclWithIssue, 3325 CheckerNameRef CheckName, 3326 StringRef name, StringRef category, 3327 StringRef str, PathDiagnosticLocation Loc, 3328 ArrayRef<SourceRange> Ranges, 3329 ArrayRef<FixItHint> Fixits) { 3330 // 'BT' is owned by BugReporter. 3331 BugType *BT = getBugTypeForName(CheckName, name, category); 3332 auto R = std::make_unique<BasicBugReport>(*BT, str, Loc); 3333 R->setDeclWithIssue(DeclWithIssue); 3334 for (const auto &SR : Ranges) 3335 R->addRange(SR); 3336 for (const auto &FH : Fixits) 3337 R->addFixItHint(FH); 3338 emitReport(std::move(R)); 3339 } 3340 3341 BugType *BugReporter::getBugTypeForName(CheckerNameRef CheckName, 3342 StringRef name, StringRef category) { 3343 SmallString<136> fullDesc; 3344 llvm::raw_svector_ostream(fullDesc) << CheckName.getName() << ":" << name 3345 << ":" << category; 3346 std::unique_ptr<BugType> &BT = StrBugTypes[fullDesc]; 3347 if (!BT) 3348 BT = std::make_unique<BugType>(CheckName, name, category); 3349 return BT.get(); 3350 } 3351